{ "site": "https://thecontinuitylayer.com", "publisher": "The Continuity Layer", "file": "https://thecontinuitylayer.com/open-data/shelves.json", "about": "https://thecontinuitylayer.com/open-data", "license": "CC BY 4.0", "licenseUrl": "https://creativecommons.org/licenses/by/4.0/", "attribution": "The Continuity Layer, https://thecontinuitylayer.com", "generated": "2026-11-09T16:53:12.392Z", "method": "Graded from the providers' own pages: each cell is read from the maker's documentation, graded yes, partly or no, dated, with the source page. No provider pays for a grade.", "grades": { "yes": 1, "partly": 0.5, "no": 0, "not stated": 0 }, "note": "A \"not stated\" cell counts for nothing and is reported in notStated; for a test marked invert, no is the good answer and scores full.", "shelves": [ { "slug": "coding-agents", "name": "Coding agents", "page": "https://thecontinuitylayer.com/coding-agents", "checked": "2026-11-03", "tests": [ { "id": "C1", "short": "Remembers", "question": "Does {name} remember across sessions?", "yes": "Persistent instructions or stored memory carry into a new session without you pasting them again." }, { "id": "C2", "short": "See it", "question": "Can you see what {name} remembered?", "yes": "A file or panel lists the stored memory." }, { "id": "C3", "short": "Edit it", "question": "Can you edit {name}'s memory?", "yes": "The memory is editable by you." }, { "id": "C4", "short": "Delete it", "question": "Can you delete or turn off {name}'s memory?", "yes": "A documented way to delete it or switch it off." }, { "id": "C5", "short": "Your file", "question": "Does {name} keep memory in a file you own?", "yes": "The memory lives in a plain file on your machine or in your repository." }, { "id": "C6", "short": "Portable", "question": "Does {name} read a format other tools read?", "yes": "It reads a shared convention such as AGENTS.md, so the same file works elsewhere." }, { "id": "C7", "short": "Training", "question": "Does {name} use your code for training by default?", "yes": "Here no is the good answer: the maker's page says your data is not used for training. Yes means it is unless you opt out; partly means it depends on the plan or the page does not give the default.", "invert": true }, { "id": "C8", "short": "Price", "question": "What does {name} cost?", "yes": "A free tier exists and the entry paid price is published." } ], "entries": [ { "slug": "claude-code", "product": "Claude Code", "maker": "Anthropic", "what": "Anthropic's agentic coding tool: reads your codebase, edits files and runs commands from the terminal, an IDE, the desktop app or the browser.", "page": "https://thecontinuitylayer.com/coding-agents/claude-code", "docsUrl": "https://code.claude.com/docs/en/overview", "pricingUrl": "https://claude.com/pricing", "license": "Not stated in the docs; no open-source license named.", "kind": null, "score": 7.5, "outOf": 8, "notStated": 0, "cells": [ { "testId": "C1", "grade": "yes", "says": "Two mechanisms carry knowledge across sessions: CLAUDE.md files you write, and auto memory notes Claude writes itself; both load at every session start.", "source": "https://code.claude.com/docs/en/memory.md", "checked": "2026-11-03" }, { "testId": "C2", "grade": "yes", "says": "/memory lists the loaded CLAUDE.md, CLAUDE.local.md and rules files and opens the auto-memory folder; auto memory is a MEMORY.md index plus topic files you can read.", "source": "https://code.claude.com/docs/en/memory.md", "checked": "2026-11-03" }, { "testId": "C3", "grade": "yes", "says": "Auto memory files are plain Markdown you can edit or delete at any time; /memory opens any of them in your editor.", "source": "https://code.claude.com/docs/en/memory.md", "checked": "2026-11-03" }, { "testId": "C4", "grade": "yes", "says": "Auto memory is on by default locally and can be turned off through /memory, the autoMemoryEnabled setting or an environment variable; the files can be deleted.", "source": "https://code.claude.com/docs/en/memory.md", "checked": "2026-11-03" }, { "testId": "C5", "grade": "yes", "says": "CLAUDE.md lives in the repository or your home folder; auto memory lives under ~/.claude/projects//memory/ and stays on your machine, not synced.", "source": "https://code.claude.com/docs/en/memory.md", "checked": "2026-11-03" }, { "testId": "C6", "grade": "yes", "says": "Reads AGENTS.md directly as project instructions when no CLAUDE.md is present; /init and /import can bring in .cursorrules, .windsurfrules and .clinerules.", "source": "https://code.claude.com/docs/en/memory.md", "checked": "2026-11-03" }, { "testId": "C7", "grade": "partly", "says": "Consumer plans (Free, Pro, Max, including Claude Code) can be used for training when the \"Help improve Claude\" setting is on; the page does not state the default. Commercial plans and the API are not used for training unless opted in. Opt out in claude.ai Settings, Privacy.", "source": "https://code.claude.com/docs/en/data-usage.md", "checked": "2026-11-03" }, { "testId": "C8", "grade": "yes", "says": "A free plan exists but does not include Claude Code; Pro at $20 a month is the entry tier with Claude Code, then Max at $100 and Ultra at $200.", "source": "https://claude.com/pricing", "checked": "2026-11-03" } ] }, { "slug": "openai-codex", "product": "OpenAI Codex", "maker": "OpenAI", "what": "OpenAI's coding agent, run locally through the CLI, an IDE or the desktop app, and in the cloud through ChatGPT.", "page": "https://thecontinuitylayer.com/coding-agents/openai-codex", "docsUrl": "https://learn.chatgpt.com/docs", "pricingUrl": "https://learn.chatgpt.com/docs/pricing", "license": "Apache-2.0 for the Codex CLI repository; the cloud service is proprietary.", "kind": null, "score": 7, "outOf": 8, "notStated": 0, "cells": [ { "testId": "C1", "grade": "yes", "says": "AGENTS.md files (global, repository and subfolder) load automatically; local memories carry context from earlier chats into later work until reset.", "source": "https://learn.chatgpt.com/docs/customization/memories.md", "checked": "2026-11-03" }, { "testId": "C2", "grade": "yes", "says": "Local memories are files under ~/.codex/memories/ you can open; AGENTS.md is a plain file.", "source": "https://learn.chatgpt.com/docs/customization/memories.md", "checked": "2026-11-03" }, { "testId": "C3", "grade": "partly", "says": "AGENTS.md is yours to edit; the memory files are generated state, and the docs say not to rely on editing them by hand.", "source": "https://learn.chatgpt.com/docs/customization/memories.md", "checked": "2026-11-03" }, { "testId": "C4", "grade": "yes", "says": "Local memories are off by default; the app's Personalization settings have a control to delete Codex memories, and there is a per-chat /memories toggle.", "source": "https://learn.chatgpt.com/docs/customization/memories.md", "checked": "2026-11-03" }, { "testId": "C5", "grade": "yes", "says": "AGENTS.md and ~/.codex/config.toml are plain files; memories are stored as local files under ~/.codex/memories/.", "source": "https://learn.chatgpt.com/docs/customization/memories.md", "checked": "2026-11-03" }, { "testId": "C6", "grade": "yes", "says": "Uses AGENTS.md, described as an open-format README for agents; the docs do not mention reading CLAUDE.md or .cursorrules.", "source": "https://developers.openai.com/codex/learn/best-practices.md", "checked": "2026-11-03" }, { "testId": "C7", "grade": "partly", "says": "The pricing page lists \"no training on your data by default\" for Business, Enterprise and API key use, and not for Plus or Pro; the consumer opt-out article could not be read on the day checked.", "source": "https://learn.chatgpt.com/docs/pricing.md", "checked": "2026-11-03" }, { "testId": "C8", "grade": "yes", "says": "Free and Go ($8) include Codex in the desktop app only; Plus at $20 a month is the entry tier with the CLI, IDE, web and cloud; Pro from $100; Business $20 to $25 a user.", "source": "https://learn.chatgpt.com/docs/pricing.md", "checked": "2026-11-03" } ] }, { "slug": "cline", "product": "Cline", "maker": "Cline Bot Inc.", "what": "An open-source coding agent in your IDE, terminal and desktop, with rules and a documented Memory Bank pattern for carrying context between sessions.", "page": "https://thecontinuitylayer.com/coding-agents/cline", "docsUrl": "https://docs.cline.bot", "pricingUrl": "https://cline.bot/pricing", "license": "Apache-2.0, per the repository.", "kind": null, "score": 7, "outOf": 8, "notStated": 1, "cells": [ { "testId": "C1", "grade": "yes", "says": "Cline Rules give persistent instructions across all conversations, set once and applied in new tasks; the Memory Bank is a documented rules pattern of Markdown files, because Cline's memory otherwise resets between sessions.", "source": "https://docs.cline.bot/customization/memory-bank", "checked": "2026-11-03" }, { "testId": "C2", "grade": "yes", "says": "The Rules panel lists project and global rules; Memory Bank files are ordinary Markdown in your project.", "source": "https://docs.cline.bot/customization/cline-rules.md", "checked": "2026-11-03" }, { "testId": "C3", "grade": "yes", "says": "Rules are plain Markdown files you create and edit; Memory Bank files are editable the same way.", "source": "https://docs.cline.bot/customization/cline-rules.md", "checked": "2026-11-03" }, { "testId": "C4", "grade": "yes", "says": "Individual rules can be toggled off without deleting them, files can be removed, and the Memory Bank is only active if you add its instructions.", "source": "https://docs.cline.bot/customization/cline-rules.md", "checked": "2026-11-03" }, { "testId": "C5", "grade": "yes", "says": "Rules live in .clinerules/ or .cline/rules/ in the project and in a global rules folder; the Memory Bank is a folder in the project.", "source": "https://docs.cline.bot/customization/cline-rules.md", "checked": "2026-11-03" }, { "testId": "C6", "grade": "yes", "says": "Reads AGENTS.md in the project and ~/.agents/AGENTS.md as a cross-tool standard; also detects .cursorrules and .windsurfrules, toggleable.", "source": "https://docs.cline.bot/customization/cline-rules.md", "checked": "2026-11-03" }, { "testId": "C7", "grade": "not stated", "says": "Telemetry is on by default with an opt-out and excludes code and prompts except for Cline's own providers; no page read states whether any data trains models.", "source": "https://docs.cline.bot/more-info/telemetry", "checked": "2026-11-03" }, { "testId": "C8", "grade": "yes", "says": "Free for individuals as an open-source extension; you pay only for the model you use, at cost or with your own key. No paid personal tier.", "source": "https://cline.bot/pricing", "checked": "2026-11-03" } ] }, { "slug": "windsurf", "product": "Windsurf (now Devin Desktop)", "maker": "Cognition", "what": "The AI editor formerly called Windsurf, whose docs now describe it as Devin Desktop with the Devin Local agent; the old docs address redirects there. Rules and AGENTS.md are injected into every session.", "page": "https://thecontinuitylayer.com/coding-agents/windsurf", "docsUrl": "https://docs.devin.ai/desktop", "pricingUrl": "https://docs.devin.ai/admin/billing/self-serve", "license": "Proprietary; no license stated in the docs.", "kind": null, "score": 6.5, "outOf": 8, "notStated": 1, "cells": [ { "testId": "C1", "grade": "yes", "says": "Always-on rules (AGENTS.md, .devin/rules, global rules) are injected at the start of every session; the older Cascade page says memories and rules carry context across conversations.", "source": "https://docs.devin.ai/desktop/devin-local/customizations", "checked": "2026-11-03" }, { "testId": "C2", "grade": "yes", "says": "The Customize page's Rules tab lists the loaded rules; the older page says memories are viewable in the Customizations panel.", "source": "https://docs.devin.ai/desktop/devin-local/customizations", "checked": "2026-11-03" }, { "testId": "C3", "grade": "yes", "says": "Rule files are edited directly; the older page says a memory can be opened and edited.", "source": "https://docs.windsurf.com/ja/windsurf/cascade/memories.md", "checked": "2026-11-03" }, { "testId": "C4", "grade": "partly", "says": "Rules are files you can remove; neither the current nor the older page describes deleting auto-generated memories or turning their generation off.", "source": "https://docs.devin.ai/desktop/devin-local/customizations", "checked": "2026-11-03" }, { "testId": "C5", "grade": "yes", "says": "Rules live in .devin/rules/*.md, .devin/global_rules.md and ~/.devin/rules, plus AGENTS.md; the older page puts memories under ~/.codeium/windsurf/memories/.", "source": "https://docs.devin.ai/desktop/devin-local/customizations", "checked": "2026-11-03" }, { "testId": "C6", "grade": "yes", "says": "AGENTS.md is read natively (root always on, subfolders scoped); legacy .windsurf/rules and .windsurfrules still load; CLAUDE.md is not mentioned.", "source": "https://docs.devin.ai/desktop/devin-local/customizations", "checked": "2026-11-03" }, { "testId": "C7", "grade": "not stated", "says": "Two official pages disagree on the day checked: the admin security page says data may be used for training by default with an opt-out in Data Controls on paid plans; the security page says training happens only if you opt in. Graded not stated until the pages agree.", "source": "https://docs.devin.ai/admin/security", "checked": "2026-11-03" }, { "testId": "C8", "grade": "yes", "says": "A free plan exists; Pro at $20 a month is the entry paid plan, then Max $200; Teams $40 a seat.", "source": "https://docs.devin.ai/admin/billing/self-serve.md", "checked": "2026-11-03" } ] }, { "slug": "gemini-cli", "product": "Gemini CLI", "maker": "Google", "what": "Google's open-source agent that brings Gemini into the terminal, with GEMINI.md context files and saved memories. Google-account sign-in for individuals ended on 18 June 2026; API-key use continues.", "page": "https://thecontinuitylayer.com/coding-agents/gemini-cli", "docsUrl": "https://geminicli.com/docs", "pricingUrl": "https://geminicli.com/docs/resources/quota-and-pricing/", "license": "Apache-2.0, stated on the Gemini CLI terms page.", "kind": null, "score": 6.5, "outOf": 8, "notStated": 1, "cells": [ { "testId": "C1", "grade": "yes", "says": "A global ~/.gemini/GEMINI.md and project GEMINI.md files are sent with every prompt; \"remember\" requests write to a memory Markdown file that loads in all future sessions.", "source": "https://geminicli.com/docs/cli/gemini-md.md", "checked": "2026-11-03" }, { "testId": "C2", "grade": "yes", "says": "/memory show prints the combined loaded context, /memory list shows the file paths, and /memory inbox shows auto-memory proposals.", "source": "https://geminicli.com/docs/reference/commands", "checked": "2026-11-03" }, { "testId": "C3", "grade": "yes", "says": "Memories are Markdown files you can edit by hand or through the tool; /memory refresh reloads them.", "source": "https://geminicli.com/docs/tools/memory", "checked": "2026-11-03" }, { "testId": "C4", "grade": "yes", "says": "The files are yours to remove; auto memory is experimental and off by default, and its proposals never apply without review.", "source": "https://geminicli.com/docs/cli/auto-memory.md", "checked": "2026-11-03" }, { "testId": "C5", "grade": "yes", "says": "Everything lives in ~/.gemini/GEMINI.md, the repository's GEMINI.md files and a per-project memory folder, all plain Markdown.", "source": "https://geminicli.com/docs/tools/memory", "checked": "2026-11-03" }, { "testId": "C6", "grade": "yes", "says": "The context file name is a setting that accepts one or more names such as AGENTS.md, CONTEXT.md or GEMINI.md.", "source": "https://geminicli.com/docs/cli/gemini-md.md", "checked": "2026-11-03" }, { "testId": "C7", "grade": "partly", "says": "The CLI's own terms page defers to the terms of whichever sign-in you use. Under the Gemini API terms, unpaid use may be used to improve Google's products and paid use is not. The individual Google-account tiers ended on 18 June 2026.", "source": "https://ai.google.dev/gemini-api/terms", "checked": "2026-11-03" }, { "testId": "C8", "grade": "not stated", "says": "The free Google-account tier and the Google AI Pro and Ultra access ended on 18 June 2026, with users pointed to Antigravity CLI; the unpaid API-key tier remains. No subscription price in US dollars appears on the Gemini CLI pricing page.", "source": "https://developers.google.com/gemini-code-assist/docs/deprecations/code-assist-individuals", "checked": "2026-11-03" } ] }, { "slug": "aider", "product": "Aider", "maker": "Aider-AI", "what": "AI pair programming in your terminal, on a new or existing codebase, with any model you bring.", "page": "https://thecontinuitylayer.com/coding-agents/aider", "docsUrl": "https://aider.chat/docs/", "pricingUrl": "https://aider.chat/docs/", "license": "Apache-2.0, per the repository.", "kind": null, "score": 6.5, "outOf": 8, "notStated": 1, "cells": [ { "testId": "C1", "grade": "partly", "says": "A conventions file listed under read: in .aider.conf.yml loads every session; earlier chat messages come back only if restore-chat-history is turned on, which is off by default.", "source": "https://aider.chat/docs/config/aider_conf.html", "checked": "2026-11-03" }, { "testId": "C2", "grade": "yes", "says": "Everything is a local file: CONVENTIONS.md, .aider.chat.history.md, .aider.input.history and an optional .aider.llm.history.", "source": "https://aider.chat/docs/config/aider_conf.html", "checked": "2026-11-03" }, { "testId": "C3", "grade": "yes", "says": "The conventions file is plain Markdown you write; the history files are plain text.", "source": "https://aider.chat/docs/usage/conventions.html", "checked": "2026-11-03" }, { "testId": "C4", "grade": "yes", "says": "Remove the read: entry or the file; chat history is not re-injected unless you turn restore-chat-history on.", "source": "https://aider.chat/docs/config/aider_conf.html", "checked": "2026-11-03" }, { "testId": "C5", "grade": "yes", "says": "All context and history lives in files in your repository or home folder.", "source": "https://aider.chat/docs/config/aider_conf.html", "checked": "2026-11-03" }, { "testId": "C6", "grade": "not stated", "says": "The docs show reading any Markdown file with /read or --read, but do not mention AGENTS.md or CLAUDE.md.", "source": "https://aider.chat/docs/usage/conventions.html", "checked": "2026-11-03" }, { "testId": "C7", "grade": "no", "says": "Aider never collects your code, chat messages, keys or personal information; analytics are opt-in. Your model provider's terms are separate.", "source": "https://aider.chat/docs/legal/privacy.html", "checked": "2026-11-03" }, { "testId": "C8", "grade": "yes", "says": "Free and open source with no paid tier or pricing page; the cost is whatever your chosen model provider charges.", "source": "https://aider.chat/docs/", "checked": "2026-11-03" } ] }, { "slug": "cursor", "product": "Cursor", "maker": "Anysphere", "what": "An AI code editor with an Agent chat, Tab completions and rules: persistent instructions the Agent follows in every conversation.", "page": "https://thecontinuitylayer.com/coding-agents/cursor", "docsUrl": "https://cursor.com/docs", "pricingUrl": "https://cursor.com/pricing", "license": "Proprietary; no license stated in the docs.", "kind": null, "score": 6, "outOf": 8, "notStated": 0, "cells": [ { "testId": "C1", "grade": "yes", "says": "Rules are persistent instructions so you do not repeat yourself in each conversation; Memories, auto-generated rules, keep context across sessions, scoped to the project.", "source": "https://cursor.com/help/customization/rules.md", "checked": "2026-11-03" }, { "testId": "C2", "grade": "yes", "says": "Project rules are files in .cursor/rules/; Memories are managed from Cursor Settings, Rules.", "source": "https://docs.cursor.com/context/memories", "checked": "2026-11-03" }, { "testId": "C3", "grade": "partly", "says": "Rules are Markdown files you create and edit; the Memories page does not describe editing an individual memory.", "source": "https://docs.cursor.com/context/memories", "checked": "2026-11-03" }, { "testId": "C4", "grade": "partly", "says": "A rule is removed by deleting its file and team rules can be turned off; the Memories page does not say how to delete or turn off memories.", "source": "https://docs.cursor.com/context/memories", "checked": "2026-11-03" }, { "testId": "C5", "grade": "partly", "says": "Project rules are versioned files in .cursor/rules/; user rules are stored on your Cursor account and left out of profile exports; where memories are stored is not stated.", "source": "https://cursor.com/help/customization/rules.md", "checked": "2026-11-03" }, { "testId": "C6", "grade": "yes", "says": "An AGENTS.md in the project root is picked up automatically, and a CLAUDE.md is read the same way.", "source": "https://cursor.com/help/customization/rules.md", "checked": "2026-11-03" }, { "testId": "C7", "grade": "partly", "says": "With Privacy Mode off, Cursor may store and use code, prompts and editor actions to train models; with it on, nothing is used for training. It is on by default for Teams and Enterprise; the default for individual plans is not stated. Opt out in Settings, General, Privacy Mode.", "source": "https://cursor.com/help/security-and-privacy/privacy.md", "checked": "2026-11-03" }, { "testId": "C8", "grade": "yes", "says": "A free plan exists; the entry paid tier is Plus at $8 a month, then Premium $30, Super $100 and Ultra $200. The older Pro at $20 stays only for existing subscribers.", "source": "https://cursor.com/help/account-and-billing/pricing", "checked": "2026-11-03" } ] }, { "slug": "github-copilot", "product": "GitHub Copilot", "maker": "GitHub (Microsoft)", "what": "GitHub's assistant and coding agent: it can research a repository, plan changes and implement them in the background; Copilot Memory stores repository facts and your preferences.", "page": "https://thecontinuitylayer.com/coding-agents/github-copilot", "docsUrl": "https://docs.github.com/en/copilot", "pricingUrl": "https://docs.github.com/en/copilot/concepts/billing/individual-plans", "license": "Proprietary; no license stated in the docs.", "kind": null, "score": 6, "outOf": 8, "notStated": 0, "cells": [ { "testId": "C1", "grade": "yes", "says": "Repository instructions (.github/copilot-instructions.md, AGENTS.md) apply to every request; Copilot Memory carries facts into later interactions (public preview, paid plans).", "source": "https://docs.github.com/en/copilot/concepts/agents/copilot-memory", "checked": "2026-11-03" }, { "testId": "C2", "grade": "yes", "says": "You can view your memories under your profile's Copilot settings; repository owners see repository facts in repository settings; instruction files are visible in the repository and in the chat's References list.", "source": "https://docs.github.com/en/copilot/how-tos/use-copilot-agents/copilot-memory/manage-for-yourself", "checked": "2026-11-03" }, { "testId": "C3", "grade": "partly", "says": "Instruction files are editable Markdown; the memory pages describe viewing and deleting memories but not editing them.", "source": "https://docs.github.com/en/copilot/how-tos/use-copilot-agents/copilot-memory/manage-for-yourself", "checked": "2026-11-03" }, { "testId": "C4", "grade": "yes", "says": "Memories can be deleted one at a time or in bulk; the feature can be turned off in Copilot settings or with /memory off; unused entries expire after 28 days.", "source": "https://docs.github.com/en/copilot/how-tos/use-copilot-agents/copilot-memory/manage-for-yourself", "checked": "2026-11-03" }, { "testId": "C5", "grade": "partly", "says": "Instructions are plain files in your repository; Copilot Memory is held by GitHub, and export is described only for Business and Enterprise administrators, not individuals.", "source": "https://docs.github.com/en/copilot/concepts/prompting-and-context/copilot-memory", "checked": "2026-11-03" }, { "testId": "C6", "grade": "yes", "says": "Reads AGENTS.md anywhere in the repository (nearest wins), or a single root CLAUDE.md or GEMINI.md instead.", "source": "https://docs.github.com/en/copilot/how-tos/configure-custom-instructions/add-repository-instructions", "checked": "2026-11-03" }, { "testId": "C7", "grade": "yes", "says": "The setting \"Allow GitHub to use my data for AI model training\" on Free, Pro, Pro+ and Max covers inputs, outputs, code snippets and context; opt out by setting it to Disabled in Copilot settings. Business and Enterprise are excluded.", "source": "https://docs.github.com/en/copilot/how-tos/manage-your-account/manage-policies", "checked": "2026-11-03" }, { "testId": "C8", "grade": "yes", "says": "Copilot Free exists with limited chat and agent use; Pro at $10 a month is the entry paid tier, then Pro+ $39 and Max $100. Copilot Memory needs a paid plan.", "source": "https://docs.github.com/en/copilot/concepts/billing/individual-plans", "checked": "2026-11-03" } ] } ] }, { "slug": "memory-systems", "name": "Memory systems", "page": "https://thecontinuitylayer.com/memory-systems", "checked": "2026-11-03", "tests": [ { "id": "M1", "short": "Open source", "question": "Is {name} open source?", "yes": "The core is under an open-source license." }, { "id": "M2", "short": "Self-host", "question": "Can you run {name} yourself?", "yes": "It runs on your own machine or server." }, { "id": "M3", "short": "Export", "question": "Can you take your memory out of {name}?", "yes": "An export, or the data sits in a standard store you own." }, { "id": "M4", "short": "Any model", "question": "Does {name} work with more than one model provider?", "yes": "It is not tied to one provider's models." }, { "id": "M5", "short": "Hosted", "question": "Is there a hosted {name}?", "yes": "A cloud or managed option exists." }, { "id": "M6", "short": "Free tier", "question": "Does {name} have a free tier, and what is the entry price?", "yes": "A free tier exists and the entry paid price is published." }, { "id": "M7", "short": "Scoping", "question": "Does {name} scope memory per user and per session?", "yes": "Both per-user and per-session or per-agent memory are supported." }, { "id": "M8", "short": "Delete", "question": "Can you delete a user's memories from {name}?", "yes": "A documented way to delete everything for one user." } ], "entries": [ { "slug": "mem0", "product": "Mem0", "maker": "Mem0", "kind": "Memory API", "what": "A memory layer for AI agents: an open-source library and server, plus a hosted platform, that extracts, stores and retrieves memories per user, agent and session.", "page": "https://thecontinuitylayer.com/memory-systems/mem0", "docsUrl": "https://docs.mem0.ai/", "pricingUrl": "https://www.mem0.ai/pricing", "repoUrl": "https://github.com/mem0ai/mem0", "license": "Apache-2.0", "status": "Active", "score": 7.5, "outOf": 8, "notStated": 0, "cells": [ { "testId": "M1", "grade": "yes", "says": "The repository is licensed Apache-2.0.", "source": "https://github.com/mem0ai/mem0", "checked": "2026-11-03" }, { "testId": "M2", "grade": "yes", "says": "The open-source version is described as the same memory engine as the platform, running on your own infrastructure as a library or a Docker Compose server.", "source": "https://docs.mem0.ai/open-source/overview", "checked": "2026-11-03" }, { "testId": "M3", "grade": "yes", "says": "Self-hosted memory sits in a vector database you run; the platform adds a Memory Export API that returns memories shaped to a schema you supply, filtered by user, agent or run.", "source": "https://docs.mem0.ai/platform/features/memory-export", "checked": "2026-11-03" }, { "testId": "M4", "grade": "yes", "says": "The model list includes OpenAI, Anthropic, Google, Mistral, DeepSeek, xAI, AWS Bedrock, Azure, Groq, Together, Ollama, LM Studio and more.", "source": "https://docs.mem0.ai/components/llms/overview", "checked": "2026-11-03" }, { "testId": "M5", "grade": "yes", "says": "A hosted platform with managed infrastructure and a dashboard.", "source": "https://docs.mem0.ai/platform/platform-vs-oss.md", "checked": "2026-11-03" }, { "testId": "M6", "grade": "partly", "says": "A free plan with monthly limits exists; beyond it the page shows usage-based overage and a custom plan, with the older Starter, Growth and Pro plans no longer sold to new accounts. No flat paid price is published.", "source": "https://www.mem0.ai/pricing", "checked": "2026-11-03" }, { "testId": "M7", "grade": "yes", "says": "Memories are scoped by user_id, agent_id, app_id and run_id, where a run is a short-lived flow, ticket or session thread; any combination works on add and as a filter.", "source": "https://docs.mem0.ai/platform/features/entity-scoped-memory.md", "checked": "2026-11-03" }, { "testId": "M8", "grade": "yes", "says": "A delete endpoint removes memories by filter (user, agent, app, run or metadata); at least one filter is required.", "source": "https://docs.mem0.ai/api-reference/memory/delete-memories.md", "checked": "2026-11-03" } ] }, { "slug": "langmem", "product": "LangMem and LangGraph Store", "maker": "LangChain", "kind": "Framework module", "what": "LangGraph's Store (namespaced JSON documents in a database you choose) plus the LangMem toolkit that extracts and manages memories in it.", "page": "https://thecontinuitylayer.com/memory-systems/langmem", "docsUrl": "https://docs.langchain.com/oss/python/langchain/long-term-memory", "pricingUrl": "https://www.langchain.com/langgraph-platform-pricing", "repoUrl": "https://github.com/langchain-ai/langmem", "license": "MIT", "status": "Active", "score": 7.5, "outOf": 8, "notStated": 0, "cells": [ { "testId": "M1", "grade": "yes", "says": "The langmem repository is MIT, and its tools work in any LangGraph app.", "source": "https://github.com/langchain-ai/langmem", "checked": "2026-11-03" }, { "testId": "M2", "grade": "yes", "says": "A pip-installed library on top of a store you run: in-memory for development, Postgres, Redis, MongoDB or Upstash in production.", "source": "https://docs.langchain.com/oss/python/langgraph/stores.md", "checked": "2026-11-03" }, { "testId": "M3", "grade": "yes", "says": "Memories are saved as JSON documents under a namespace and key in a store you run.", "source": "https://docs.langchain.com/oss/python/langchain/long-term-memory", "checked": "2026-11-03" }, { "testId": "M4", "grade": "yes", "says": "Examples use Google, OpenAI, Anthropic, OpenRouter, Fireworks, Baseten and Ollama.", "source": "https://docs.langchain.com/oss/python/langchain/long-term-memory", "checked": "2026-11-03" }, { "testId": "M5", "grade": "yes", "says": "The store is available by default in LangGraph Platform deployments, where the hosted server handles storage.", "source": "https://langchain-ai.github.io/langmem", "checked": "2026-11-03" }, { "testId": "M6", "grade": "yes", "says": "The library is free; LangGraph Platform has a Developer plan at $0 a seat with pay-as-you-go, Plus at $39 a seat a month, and Enterprise custom.", "source": "https://www.langchain.com/langgraph-platform-pricing", "checked": "2026-11-03" }, { "testId": "M7", "grade": "yes", "says": "Namespaces are tuples such as (user_id, \"memories\"); threads are scoped by the checkpointer for short-term memory while the store holds memory across threads.", "source": "https://docs.langchain.com/oss/python/langgraph/stores.md", "checked": "2026-11-03" }, { "testId": "M8", "grade": "partly", "says": "The store interface deletes one item at a time by namespace and key; no per-user bulk delete is documented.", "source": "https://docs.langchain.com/oss/python/langgraph/stores.md", "checked": "2026-11-03" } ] }, { "slug": "memos", "product": "MemOS", "maker": "MemTensor", "kind": "Memory API", "what": "A self-described memory operating system for agents with composable memory cubes, hybrid retrieval and skill memory; an open-source server plus a cloud API.", "page": "https://thecontinuitylayer.com/memory-systems/memos", "docsUrl": "https://memos-docs.openmem.net/", "pricingUrl": "https://memos-docs.openmem.net/memos_cloud/limit", "repoUrl": "https://github.com/MemTensor/MemOS", "license": "Apache-2.0", "status": "Active", "score": 7, "outOf": 8, "notStated": 0, "cells": [ { "testId": "M1", "grade": "yes", "says": "The repository is Apache-2.0, and the open-source edition runs entirely in your own environment.", "source": "https://github.com/MemTensor/MemOS", "checked": "2026-11-03" }, { "testId": "M2", "grade": "yes", "says": "Self-hosted with Docker Compose (Neo4j and Qdrant required) or directly; a local plugin uses SQLite.", "source": "https://github.com/MemTensor/MemOS", "checked": "2026-11-03" }, { "testId": "M3", "grade": "partly", "says": "Self-hosted memory sits in Neo4j and Qdrant that you run and is described as inspectable and editable; no export or dump is documented.", "source": "https://github.com/MemTensor/MemOS", "checked": "2026-11-03" }, { "testId": "M4", "grade": "yes", "says": "Model backends include Ollama, OpenAI or an OpenAI-compatible proxy, Qwen, DeepSeek and Hugging Face; Anthropic and Gemini are not listed.", "source": "https://memos-docs.openmem.net/open_source/modules/model_backend", "checked": "2026-11-03" }, { "testId": "M5", "grade": "yes", "says": "A fully managed cloud API with a console.", "source": "https://memos-docs.openmem.net/memos_cloud/cloud_and_opensource", "checked": "2026-11-03" }, { "testId": "M6", "grade": "partly", "says": "The cloud offers free trial quotas per developer account; no price in US dollars appears on any page checked.", "source": "https://memos-docs.openmem.net/memos_cloud/limit", "checked": "2026-11-03" }, { "testId": "M7", "grade": "yes", "says": "Requests carry a user id; memory cubes isolate or share memory across users, projects and agents; the cloud add call takes a conversation id.", "source": "https://github.com/MemTensor/MemOS", "checked": "2026-11-03" }, { "testId": "M8", "grade": "yes", "says": "The cloud delete call with only a user id clears all of that user's memories in the project; the self-hosted server has a delete-memories-for-user endpoint.", "source": "https://memos-docs.openmem.net/memos_cloud/mem_operations/delete_memory", "checked": "2026-11-03" } ] }, { "slug": "letta", "product": "Letta", "maker": "Letta (formerly MemGPT)", "kind": "Agent harness", "what": "A platform for stateful agents with editable memory blocks, archival memory and portable Agent Files, run as Letta Cloud or a self-hosted server.", "page": "https://thecontinuitylayer.com/memory-systems/letta", "docsUrl": "https://docs.letta.com/", "pricingUrl": "https://docs.letta.com/pricing", "repoUrl": "https://github.com/letta-ai/letta", "license": "Apache-2.0", "status": "Active", "score": 6.5, "outOf": 8, "notStated": 0, "cells": [ { "testId": "M1", "grade": "yes", "says": "Both the letta and letta-code repositories are Apache-2.0; the README says active development is in letta-code.", "source": "https://github.com/letta-ai/letta", "checked": "2026-11-03" }, { "testId": "M2", "grade": "yes", "says": "A self-hosted server is documented, with Docker, Docker Compose, Railway and Fly.io configurations.", "source": "https://docs.letta.com/self-hosting", "checked": "2026-11-03" }, { "testId": "M3", "grade": "partly", "says": "An Agent File export contains the state needed to recreate the same agent, including memory blocks and message history; archival passages are marked as planned and secrets are exported empty.", "source": "https://docs.letta.com/guides/core-concepts/agent-file/index.md", "checked": "2026-11-03" }, { "testId": "M4", "grade": "yes", "says": "The API supports the major providers, including Anthropic, OpenAI, Google, Azure, AWS Bedrock, OpenRouter and Ollama; agents keep their memory when the model is switched.", "source": "https://docs.letta.com/guides/build-with-letta/models", "checked": "2026-11-03" }, { "testId": "M5", "grade": "yes", "says": "Letta Cloud keeps an agent's memory, identity and conversations available across machines and is the default.", "source": "https://github.com/letta-ai/letta-code", "checked": "2026-11-03" }, { "testId": "M6", "grade": "yes", "says": "Free at $0 with your own keys and limited agents; Pro at $20 a month; a usage-based Developer plan; Enterprise custom.", "source": "https://docs.letta.com/pricing", "checked": "2026-11-03" }, { "testId": "M7", "grade": "partly", "says": "Memory belongs to an agent; shared blocks can attach to several agents, and a documented pattern attaches a user-preferences block across agents. Per-user scoping is a pattern, not a first-class store.", "source": "https://docs.letta.com/guides/core-concepts/memory/shared-memory/index.md", "checked": "2026-11-03" }, { "testId": "M8", "grade": "partly", "says": "A single archival passage can be deleted and blocks can be detached or overwritten; there is no documented delete-everything-for-a-user call because memory is agent-scoped.", "source": "https://docs.letta.com/api/resources/agents/subresources/passages/methods/delete/index.md", "checked": "2026-11-03" } ] }, { "slug": "cognee", "product": "Cognee", "maker": "Cognee", "kind": "Graph memory", "what": "An open-source memory engine that turns documents and conversations into graph and vector memory with multi-user datasets, plus a usage-priced Cognee Cloud.", "page": "https://thecontinuitylayer.com/memory-systems/cognee", "docsUrl": "https://docs.cognee.ai/", "pricingUrl": "https://cognee.ai/pricing", "repoUrl": "https://github.com/topoteretes/cognee", "license": "Apache-2.0", "status": "Active", "score": 6.5, "outOf": 8, "notStated": 0, "cells": [ { "testId": "M1", "grade": "yes", "says": "The repository is licensed Apache-2.0.", "source": "https://github.com/topoteretes/cognee", "checked": "2026-11-03" }, { "testId": "M2", "grade": "yes", "says": "Runs locally for free with no API key required, with a Docker image and Compose for the API, UI and MCP server; free to self-host.", "source": "https://www.cognee.ai/faq", "checked": "2026-11-03" }, { "testId": "M3", "grade": "partly", "says": "Self-hosted data sits in backends you run, with per-user directories; no export API is documented, though an import format exists for moving in from Mem0, Letta, Zep and Graphiti.", "source": "https://docs.cognee.ai/core-concepts/multi-user-mode/permissions-system/overview", "checked": "2026-11-03" }, { "testId": "M4", "grade": "yes", "says": "Supported providers include OpenAI, Azure OpenAI, Gemini, Anthropic, AWS Bedrock, Groq, Ollama, LM Studio, Hugging Face, llama.cpp and any OpenAI-compatible endpoint.", "source": "https://docs.cognee.ai/setup-configuration/llm-providers", "checked": "2026-11-03" }, { "testId": "M5", "grade": "yes", "says": "Cognee Cloud is the managed option.", "source": "https://github.com/topoteretes/cognee", "checked": "2026-11-03" }, { "testId": "M6", "grade": "yes", "says": "Free at $0 with one workspace and a token allowance; Standard at $1 per million tokens plus $5 per extra workspace a month; Enterprise by contact.", "source": "https://cognee.ai/pricing", "checked": "2026-11-03" }, { "testId": "M7", "grade": "partly", "says": "Scoped by user, tenant and dataset with read, write, delete and share permissions and full data isolation; session-level memory is not described.", "source": "https://docs.cognee.ai/core-concepts/multi-user-mode/permissions-system/overview", "checked": "2026-11-03" }, { "testId": "M8", "grade": "partly", "says": "Delete operations remove items and datasets the current user may delete, including graph nodes and embeddings; there is no parameter to target another user's data.", "source": "https://docs.cognee.ai/core-concepts/main-operations/delete", "checked": "2026-11-03" } ] }, { "slug": "honcho", "product": "Honcho", "maker": "Plastic Labs", "kind": "Memory API", "what": "An open-source memory server for stateful agents built around workspaces, peers and sessions, with background reasoning that builds user models; also offered as Honcho Cloud.", "page": "https://thecontinuitylayer.com/memory-systems/honcho", "docsUrl": "https://honcho.dev/docs", "pricingUrl": "https://honcho.dev/", "repoUrl": "https://github.com/plastic-labs/honcho", "license": "AGPL-3.0", "status": "Active", "score": 6.5, "outOf": 8, "notStated": 0, "cells": [ { "testId": "M1", "grade": "yes", "says": "Open source under AGPL-3.0.", "source": "https://github.com/plastic-labs/honcho", "checked": "2026-11-03" }, { "testId": "M2", "grade": "yes", "says": "The server self-hosts with Docker Compose or manually on PostgreSQL with its vector extension; the cloud runs the same open-source server.", "source": "https://honcho.dev/docs/v3/contributing/self-hosting.md", "checked": "2026-11-03" }, { "testId": "M3", "grade": "partly", "says": "Self-hosted data lives in your Postgres or another supported vector store; the docs index has an import guide from Mem0 but no export endpoint.", "source": "https://honcho.dev/docs/llms.txt", "checked": "2026-11-03" }, { "testId": "M4", "grade": "yes", "says": "OpenAI by default, plus Anthropic, Gemini and OpenAI-compatible endpoints such as OpenRouter, Together, vLLM and Ollama; models must support tool calling.", "source": "https://honcho.dev/docs/v3/contributing/configuration.md", "checked": "2026-11-03" }, { "testId": "M5", "grade": "yes", "says": "Honcho Cloud is described as the quickest way to use it in production; an instance turns on after a payment method is added.", "source": "https://honcho.dev/docs/v3/documentation/reference/platform", "checked": "2026-11-03" }, { "testId": "M6", "grade": "partly", "says": "No free tier is stated; pricing is usage-based (ingestion per million tokens, reasoning per query), with startup credits. The prices appear on the homepage, not a pricing page.", "source": "https://honcho.dev/", "checked": "2026-11-03" }, { "testId": "M7", "grade": "yes", "says": "Workspaces isolate use cases; peers (users or agents) and sessions are many-to-many; scopes group sessions to limit recall.", "source": "https://github.com/plastic-labs/honcho", "checked": "2026-11-03" }, { "testId": "M8", "grade": "partly", "says": "Peers and individual messages cannot be deleted; clearing a user means deleting every session plus the peer's remaining conclusions, or deleting the whole workspace.", "source": "https://honcho.dev/docs/v3/documentation/features/advanced/deleting-data.md", "checked": "2026-11-03" } ] }, { "slug": "zep", "product": "Zep Cloud", "maker": "Zep", "kind": "Graph memory", "what": "A managed temporal knowledge-graph context layer for agents, with per-user graphs, threads and a context API, built on the open-source Graphiti engine.", "page": "https://thecontinuitylayer.com/memory-systems/zep", "docsUrl": "https://help.getzep.com/", "pricingUrl": "https://www.getzep.com/pricing", "repoUrl": "https://github.com/getzep/zep", "license": "Graphiti is Apache-2.0; Zep Cloud is proprietary", "status": "Active", "score": 6, "outOf": 8, "notStated": 0, "cells": [ { "testId": "M1", "grade": "partly", "says": "Graphiti, the graph engine that powers Zep, is Apache-2.0; the getzep/zep repository says it is not Zep's product and that Zep Community Edition is no longer supported.", "source": "https://github.com/getzep/zep", "checked": "2026-11-03" }, { "testId": "M2", "grade": "partly", "says": "Free, Flex and Flex Plus run on Zep Cloud only; Enterprise offers deployment to your own cloud account. Otherwise self-hosting means building on Graphiti yourself.", "source": "https://www.getzep.com/pricing", "checked": "2026-11-03" }, { "testId": "M3", "grade": "partly", "says": "Read endpoints return the nodes, episodes and edges of a user's or group's graph; no bulk export is documented.", "source": "https://help.getzep.com/v2/reading-data-from-the-graph", "checked": "2026-11-03" }, { "testId": "M4", "grade": "partly", "says": "Zep sits beside whichever chat model you use; the quickstart shows OpenAI and gives guidance for Anthropic and other providers. Which model Zep uses internally is not stated.", "source": "https://help.getzep.com/v2/quickstart.md", "checked": "2026-11-03" }, { "testId": "M5", "grade": "yes", "says": "Zep Cloud is the product; a free tier with monthly credits is offered.", "source": "https://www.getzep.com/pricing", "checked": "2026-11-03" }, { "testId": "M6", "grade": "yes", "says": "Free at $0 with 10,000 credits a month; Flex at $125 a month; Flex Plus at $375; Enterprise custom.", "source": "https://www.getzep.com/pricing", "checked": "2026-11-03" }, { "testId": "M7", "grade": "yes", "says": "A user can have many threads; facts learned in one session are available across all of that user's sessions, and group graphs are shared across users. The graph does not separate data by thread.", "source": "https://help.getzep.com/users-and-user-graphs.md", "checked": "2026-11-03" }, { "testId": "M8", "grade": "yes", "says": "Deleting a user removes the user's threads, messages and graph; edges, nodes, episodes and threads can also be deleted one at a time.", "source": "https://help.getzep.com/sdk-reference/user/delete", "checked": "2026-11-03" } ] }, { "slug": "supermemory", "product": "Supermemory", "maker": "Supermemory", "kind": "Memory API", "what": "A memory and context engine offered as a hosted API and as a local single-binary server, with container-tag scoping and automatic forgetting of contradicted facts.", "page": "https://thecontinuitylayer.com/memory-systems/supermemory", "docsUrl": "https://supermemory.ai/docs", "pricingUrl": "https://supermemory.ai/pricing", "repoUrl": "https://github.com/supermemoryai/supermemory", "license": "MIT for the SDKs and plugins; the server binary is not open source", "status": "Active", "score": 6, "outOf": 8, "notStated": 0, "cells": [ { "testId": "M1", "grade": "partly", "says": "The public repository is MIT and the SDKs, plugins and MCP server are open source, but the self-hosting docs say the server binary is not open source and is free within a lite license limit.", "source": "https://supermemory.ai/docs/self-hosting/overview", "checked": "2026-11-03" }, { "testId": "M2", "grade": "yes", "says": "The full memory API runs locally as a single binary, installed by script, npx or Docker.", "source": "https://supermemory.ai/docs/self-hosting/overview", "checked": "2026-11-03" }, { "testId": "M3", "grade": "partly", "says": "Local data lives in one directory that the docs call easy to back up or move; no export API is documented for the hosted service, and moving from local to the platform does not transfer data.", "source": "https://github.com/supermemoryai/supermemory", "checked": "2026-11-03" }, { "testId": "M4", "grade": "yes", "says": "Any model: OpenAI, Anthropic, Gemini, Groq or any OpenAI-compatible endpoint such as Ollama, LM Studio or vLLM, with local embeddings by default.", "source": "https://supermemory.ai/docs/self-hosting/overview", "checked": "2026-11-03" }, { "testId": "M5", "grade": "yes", "says": "A hosted platform, reached by changing the base URL.", "source": "https://github.com/supermemoryai/supermemory", "checked": "2026-11-03" }, { "testId": "M6", "grade": "yes", "says": "Free at $0 with monthly credits; Pro at $19 a month; Max at $100; Scale at $399 with a self-hosted option; Enterprise custom.", "source": "https://supermemory.ai/pricing", "checked": "2026-11-03" }, { "testId": "M7", "grade": "partly", "says": "Memory is scoped with projects, called container tags, such as one per user; separate session scoping is not described.", "source": "https://github.com/supermemoryai/supermemory", "checked": "2026-11-03" }, { "testId": "M8", "grade": "partly", "says": "Deleting a container tag removes all its documents and memories, for owners and admins; memories are described as marked forgotten rather than confirmed erased. There is also a forget-one-memory endpoint.", "source": "https://supermemory.ai/docs/api-reference/spaces/delete-container-tag", "checked": "2026-11-03" } ] }, { "slug": "memobase", "product": "Memobase", "maker": "memodb-io", "kind": "Memory API", "what": "User-profile memory for chatbots: per-user profiles and event memory, as a dockerized server or Memobase Cloud.", "page": "https://thecontinuitylayer.com/memory-systems/memobase", "docsUrl": "https://docs.memobase.io/", "pricingUrl": "", "repoUrl": "https://github.com/memodb-io/memobase", "license": "Apache-2.0", "status": "Unclear", "score": 5.5, "outOf": 8, "notStated": 1, "cells": [ { "testId": "M1", "grade": "yes", "says": "The repository is Apache-2.0.", "source": "https://github.com/memodb-io/memobase", "checked": "2026-11-03" }, { "testId": "M2", "grade": "yes", "says": "Fully dockerized, with server setup docs and a default local address.", "source": "https://github.com/memodb-io/memobase", "checked": "2026-11-03" }, { "testId": "M3", "grade": "partly", "says": "Profiles are retrievable as JSON and the data API can return all of a user's records; no bulk export is documented.", "source": "https://docs.memobase.io/api-reference/overview", "checked": "2026-11-03" }, { "testId": "M4", "grade": "not stated", "says": "The README shows OpenAI and Ollama tutorials; no provider list appears on the pages checked.", "source": "https://github.com/memodb-io/memobase", "checked": "2026-11-03" }, { "testId": "M5", "grade": "yes", "says": "Memobase Cloud offers a free tier for testing.", "source": "https://github.com/memodb-io/memobase", "checked": "2026-11-03" }, { "testId": "M6", "grade": "partly", "says": "A free tier is stated; no paid prices were found on official pages.", "source": "https://github.com/memodb-io/memobase", "checked": "2026-11-03" }, { "testId": "M7", "grade": "partly", "says": "Per-user by design (memory for the user, not the agent); no session scoping is described.", "source": "https://github.com/memodb-io/memobase", "checked": "2026-11-03" }, { "testId": "M8", "grade": "yes", "says": "Users, their records and individual profile entries can be deleted through the API.", "source": "https://docs.memobase.io/api-reference/overview", "checked": "2026-11-03" } ] }, { "slug": "claude-memory-tool", "product": "Claude memory tool", "maker": "Anthropic", "kind": "Platform feature", "what": "A client-side, file-based memory tool on the Claude Developer Platform: Claude issues view, create, edit and delete commands against a memory directory that your application stores.", "page": "https://thecontinuitylayer.com/memory-systems/claude-memory-tool", "docsUrl": "https://platform.claude.com/en/docs/agents-and-tools/tool-use/memory-tool", "pricingUrl": "https://platform.claude.com/docs/en/about-claude/pricing", "license": "Proprietary API feature", "status": "Active", "score": 4.5, "outOf": 8, "notStated": 0, "cells": [ { "testId": "M1", "grade": "no", "says": "A feature of the Claude API; there is no open-source memory core.", "source": "https://platform.claude.com/en/docs/agents-and-tools/tool-use/memory-tool", "checked": "2026-11-03" }, { "testId": "M2", "grade": "partly", "says": "The tool runs client-side and memory lives entirely in your application, so the storage is yours while the model is hosted.", "source": "https://platform.claude.com/en/docs/agents-and-tools/tool-use/memory-tool", "checked": "2026-11-03" }, { "testId": "M3", "grade": "yes", "says": "Memory is a directory of files that your handler maps onto a per-user directory or keys in a database.", "source": "https://platform.claude.com/en/docs/agents-and-tools/tool-use/memory-tool", "checked": "2026-11-03" }, { "testId": "M4", "grade": "no", "says": "Available on Claude 4 and later models only.", "source": "https://platform.claude.com/en/docs/agents-and-tools/tool-use/memory-tool", "checked": "2026-11-03" }, { "testId": "M5", "grade": "yes", "says": "Delivered through the hosted Claude API.", "source": "https://platform.claude.com/en/docs/agents-and-tools/tool-use/memory-tool", "checked": "2026-11-03" }, { "testId": "M6", "grade": "partly", "says": "New API users receive a small amount of free credits; there is no separate charge for the memory tool; the entry model is priced per million tokens.", "source": "https://platform.claude.com/docs/en/about-claude/pricing", "checked": "2026-11-03" }, { "testId": "M7", "grade": "partly", "says": "The docs give a per-user directory as the storage example and leave isolation to your implementation.", "source": "https://platform.claude.com/en/docs/agents-and-tools/tool-use/memory-tool", "checked": "2026-11-03" }, { "testId": "M8", "grade": "yes", "says": "The delete command removes a file or directory; the storage is yours to purge.", "source": "https://platform.claude.com/en/docs/agents-and-tools/tool-use/memory-tool", "checked": "2026-11-03" } ] }, { "slug": "graphiti", "product": "Graphiti", "maker": "Zep", "kind": "Graph memory", "what": "An open-source framework for building real-time temporal knowledge graphs for agents; the engine inside Zep, run on your own graph database.", "page": "https://thecontinuitylayer.com/memory-systems/graphiti", "docsUrl": "https://help.getzep.com/graphiti/getting-started/welcome.md", "pricingUrl": "", "repoUrl": "https://github.com/getzep/graphiti", "license": "Apache-2.0", "status": "Active", "score": 4.5, "outOf": 8, "notStated": 3, "cells": [ { "testId": "M1", "grade": "yes", "says": "The repository is licensed Apache-2.0.", "source": "https://github.com/getzep/graphiti", "checked": "2026-11-03" }, { "testId": "M2", "grade": "yes", "says": "Self-hosted only: you supply the graph database, the model and the embedder.", "source": "https://github.com/getzep/graphiti", "checked": "2026-11-03" }, { "testId": "M3", "grade": "yes", "says": "Data lives in a graph database you run: Neo4j, FalkorDB or Amazon Neptune.", "source": "https://github.com/getzep/graphiti", "checked": "2026-11-03" }, { "testId": "M4", "grade": "yes", "says": "Defaults to OpenAI, with Anthropic, Gemini and Groq supported, plus OpenAI-compatible endpoints such as Ollama and Azure OpenAI.", "source": "https://github.com/getzep/graphiti", "checked": "2026-11-03" }, { "testId": "M5", "grade": "partly", "says": "There is no hosted Graphiti; the README points to Zep as the managed service built on it.", "source": "https://github.com/getzep/graphiti", "checked": "2026-11-03" }, { "testId": "M6", "grade": "not stated", "says": "No pricing page: an open-source library where you pay your own model and database costs.", "source": "https://github.com/getzep/graphiti", "checked": "2026-11-03" }, { "testId": "M7", "grade": "not stated", "says": "The README and welcome page do not describe per-user or per-session scoping.", "source": "https://help.getzep.com/graphiti/getting-started/welcome.md", "checked": "2026-11-03" }, { "testId": "M8", "grade": "not stated", "says": "Zep's deletion page covers Zep Cloud only and says it does not document deletion in Graphiti.", "source": "https://help.getzep.com/v4/deleting-data-from-the-graph.md", "checked": "2026-11-03" } ] }, { "slug": "microsoft-agent-framework", "product": "Agent Framework memory", "maker": "Microsoft", "kind": "Framework module", "what": "The open-source successor to Semantic Kernel's memory: context providers that run before and after each turn, in-memory history providers, and a managed Foundry memory provider.", "page": "https://thecontinuitylayer.com/memory-systems/microsoft-agent-framework", "docsUrl": "https://learn.microsoft.com/agent-framework/get-started/memory", "pricingUrl": "", "repoUrl": "https://github.com/microsoft/agent-framework", "license": "MIT", "status": "Active", "score": 4, "outOf": 8, "notStated": 2, "cells": [ { "testId": "M1", "grade": "yes", "says": "The repository is MIT.", "source": "https://github.com/microsoft/agent-framework", "checked": "2026-11-03" }, { "testId": "M2", "grade": "yes", "says": "A library you run; chat history is kept in an in-memory provider or in the underlying AI service, and a custom context provider can use your own storage.", "source": "https://learn.microsoft.com/agent-framework/get-started/memory", "checked": "2026-11-03" }, { "testId": "M3", "grade": "partly", "says": "Storage depends on the provider you plug in; no export feature is described.", "source": "https://learn.microsoft.com/agent-framework/get-started/memory", "checked": "2026-11-03" }, { "testId": "M4", "grade": "partly", "says": "The README names Microsoft Foundry, Azure OpenAI and OpenAI and says more providers are being added; Anthropic is not named.", "source": "https://github.com/microsoft/agent-framework", "checked": "2026-11-03" }, { "testId": "M5", "grade": "partly", "says": "A Foundry memory provider offers managed semantic memory; the framework itself has no hosted memory service.", "source": "https://learn.microsoft.com/agent-framework/get-started/memory", "checked": "2026-11-03" }, { "testId": "M6", "grade": "not stated", "says": "The framework is free and open source; Foundry memory pricing is not on the pages checked.", "source": "https://learn.microsoft.com/agent-framework/get-started/memory", "checked": "2026-11-03" }, { "testId": "M7", "grade": "partly", "says": "Sessions share context across runs and the Foundry provider accepts a user scope; thread-level scoping is not detailed.", "source": "https://learn.microsoft.com/agent-framework/get-started/memory", "checked": "2026-11-03" }, { "testId": "M8", "grade": "not stated", "says": "The memory tutorial does not cover deleting memory; the migration guide says hosted history must be deleted through the provider's own SDK.", "source": "https://learn.microsoft.com/en-us/agent-framework/migration-guide/from-semantic-kernel", "checked": "2026-11-03" } ] }, { "slug": "google-memory-bank", "product": "Agent Engine Memory Bank", "maker": "Google Cloud", "kind": "Platform feature", "what": "A managed Google Cloud service that generates, consolidates and retrieves long-term user memories from agent sessions, keyed by a scope such as a user id.", "page": "https://thecontinuitylayer.com/memory-systems/google-memory-bank", "docsUrl": "https://docs.cloud.google.com/agent-builder/agent-engine/memory-bank/overview", "pricingUrl": "https://services.google.com/fh/files/emails/b_502770571_gemini_enterprise_agent_platform_pricing.pdf", "license": "Proprietary cloud service", "status": "Active", "score": 3.5, "outOf": 8, "notStated": 1, "cells": [ { "testId": "M1", "grade": "no", "says": "A managed Google Cloud service with no open-source core.", "source": "https://docs.cloud.google.com/agent-builder/agent-engine/memory-bank/overview", "checked": "2026-11-03" }, { "testId": "M2", "grade": "no", "says": "Runs in Google Cloud regions; no self-hosted option is described.", "source": "https://docs.cloud.google.com/agent-builder/agent-engine/memory-bank/set-up", "checked": "2026-11-03" }, { "testId": "M3", "grade": "partly", "says": "Retrieving with a scope returns all memories for a user; no bulk export is described.", "source": "https://docs.cloud.google.com/agent-builder/agent-engine/memory-bank/quickstart-api", "checked": "2026-11-03" }, { "testId": "M4", "grade": "not stated", "says": "The service uses generative models to produce memories, without naming which; the SDK can be used with frameworks other than Google's, but nothing is said about non-Google models.", "source": "https://docs.cloud.google.com/agent-builder/agent-engine/memory-bank/overview", "checked": "2026-11-03" }, { "testId": "M5", "grade": "yes", "says": "A fully managed Agent Engine service.", "source": "https://docs.cloud.google.com/agent-builder/agent-engine/memory-bank/overview", "checked": "2026-11-03" }, { "testId": "M6", "grade": "partly", "says": "No free tier is mentioned; the pricing document lists per-thousand-memories storage and retrieval rates.", "source": "https://services.google.com/fh/files/emails/b_502770571_gemini_enterprise_agent_platform_pricing.pdf", "checked": "2026-11-03" }, { "testId": "M7", "grade": "yes", "says": "The scope defaults to the session's user id and accepts custom keys such as a session id; only memories with the same scope are consolidated together.", "source": "https://docs.cloud.google.com/agent-builder/agent-engine/memory-bank/quickstart-api", "checked": "2026-11-03" }, { "testId": "M8", "grade": "partly", "says": "A memory can be deleted by name; no delete-all-for-scope call is documented on the pages checked.", "source": "https://docs.cloud.google.com/agent-builder/agent-engine/memory-bank/quickstart-api", "checked": "2026-11-03" } ] }, { "slug": "openai-conversation-state", "product": "OpenAI conversation state", "maker": "OpenAI", "kind": "Platform feature", "what": "Server-side conversation state for the Responses API: stored responses chained together, with a 30-day default retention, and durable Conversation objects reusable across sessions.", "page": "https://thecontinuitylayer.com/memory-systems/openai-conversation-state", "docsUrl": "https://developers.openai.com/api/docs/guides/conversation-state", "pricingUrl": "https://developers.openai.com/api/docs/pricing", "license": "Proprietary API feature", "status": "Active", "score": 3, "outOf": 8, "notStated": 0, "cells": [ { "testId": "M1", "grade": "no", "says": "A hosted API feature with no open-source core.", "source": "https://developers.openai.com/api/docs/guides/conversation-state", "checked": "2026-11-03" }, { "testId": "M2", "grade": "no", "says": "State is stored by OpenAI; storing is on by default and can be turned off, in which case you keep the history yourself.", "source": "https://developers.openai.com/api/docs/guides/conversation-state", "checked": "2026-11-03" }, { "testId": "M3", "grade": "partly", "says": "Stored responses can be retrieved through the API or the dashboard logs, and conversation items can be listed; no bulk export is described.", "source": "https://developers.openai.com/api/docs/guides/conversation-state", "checked": "2026-11-03" }, { "testId": "M4", "grade": "no", "says": "OpenAI models only.", "source": "https://developers.openai.com/api/docs/guides/conversation-state", "checked": "2026-11-03" }, { "testId": "M5", "grade": "yes", "says": "Hosted; the Responses API is not priced separately and tokens are billed at model rates.", "source": "https://developers.openai.com/api/docs/pricing", "checked": "2026-11-03" }, { "testId": "M6", "grade": "partly", "says": "No free model tier is stated and there is no storage charge; chained earlier inputs are billed as input tokens.", "source": "https://developers.openai.com/api/docs/pricing", "checked": "2026-11-03" }, { "testId": "M7", "grade": "partly", "says": "Conversation objects are durable and reusable across sessions and devices; no user-level memory or user identifier is described.", "source": "https://developers.openai.com/api/docs/guides/conversation-state", "checked": "2026-11-03" }, { "testId": "M8", "grade": "partly", "says": "A response can be deleted and a conversation can be deleted, but deleting a conversation does not delete its items; there is no per-user delete.", "source": "https://developers.openai.com/api/reference/resources/conversations/methods/delete/index.html", "checked": "2026-11-03" } ] } ] }, { "slug": "open-models", "name": "Open models", "page": "https://thecontinuitylayer.com/open-models", "checked": "2026-11-03", "tests": [ { "id": "O1", "short": "License", "question": "Is {name} under an open license?", "yes": "The weights are under an OSI-approved license such as Apache-2.0 or MIT. Partly means the maker's own license with conditions, such as a user cap or an acceptable-use clause; no means research or non-commercial use only." }, { "id": "O2", "short": "Download", "question": "Can you download {name}'s weights?", "yes": "The maker's page links the weights for download with nothing more than a click-through license. Partly means an access form, an approval step, or only some sizes released." }, { "id": "O3", "short": "Context", "question": "Does the maker state {name}'s context window?", "yes": "A token count for the context window is on the maker's page or model card." }, { "id": "O4", "short": "Runs locally", "question": "Does the maker say what it takes to run {name} on your own machine?", "yes": "The page states the parameter count and the memory or hardware it needs. Partly means the parameter count alone." }, { "id": "O5", "short": "Hosted price", "question": "Does the maker publish a price for its hosted {name}?", "yes": "A per-token price on the maker's own pricing page. Partly means a hosted version exists but the price sits behind sign-in or a sales call." }, { "id": "O6", "short": "Training", "question": "Does the maker train on your prompts when you use its hosted {name}?", "yes": "Here no is the good answer: the maker's page says what you send its API is not used for training. Yes means it is unless you opt out; partly means it depends on the plan or the page does not give the default.", "invert": true }, { "id": "O7", "short": "Fine-tune", "question": "Can you fine-tune {name} and keep the result?", "yes": "The license allows fine-tuning and the fine-tuned weights are yours to keep and share. Partly means conditions follow the derivative, or a hosted fine-tune you cannot download." } ], "entries": [ { "slug": "mimo-v2-6-pro", "product": "MiMo-V2.6-Pro", "maker": "Xiaomi", "what": "The flagship checkpoint of Xiaomi's MiMo-V2.6 series: a sparse mixture-of-experts model with 1.02T total and 42B active parameters that takes text, image, video and audio input, served on the Xiaomi MiMo API platform.", "page": "https://thecontinuitylayer.com/open-models/mimo-v2-6-pro", "docsUrl": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL", "pricingUrl": "https://mimo.mi.com/docs/en-US/price/pay-as-you-go", "license": "MIT, per the model card; a use-restrictions notice on the card says it does not amend the license.", "kind": null, "score": 6.5, "outOf": 7, "notStated": 0, "cells": [ { "testId": "O1", "grade": "yes", "says": "The card lists the license as MIT. A Use Restrictions section lists prohibited uses such as military applications and malicious cyber activity and says it \"does not amend the applicable model licence or impose additional licence conditions.\"", "source": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL", "checked": "2026-11-03" }, { "testId": "O2", "grade": "yes", "says": "Weights for MiMo-V2.6-Pro-RL and MiMo-V2.6-Flash-RL are on Hugging Face under XiaomiMiMo, in BF16, FP8 and other tensor types.", "source": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL", "checked": "2026-11-03" }, { "testId": "O3", "grade": "yes", "says": "The card states \"Context Length: 1M tokens.\"", "source": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL", "checked": "2026-11-03" }, { "testId": "O4", "grade": "partly", "says": "1.02T total and 42B active parameters; the card gives no GPU memory figure, shows vLLM with tensor-parallel size 8 and SGLang across two nodes, and links community quantizations for llama.cpp, Ollama and LM Studio.", "source": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL", "checked": "2026-11-03" }, { "testId": "O5", "grade": "yes", "says": "Xiaomi's pay-as-you-go page (updated 21 October 2026) lists mimo-v2.6-pro at $0.435 per million input tokens on a cache miss, $0.0036 on a cache hit, and $0.87 per million output tokens; mimo-v2.6-flash at $0.14 in and $0.28 out.", "source": "https://mimo.mi.com/docs/en-US/price/pay-as-you-go", "checked": "2026-11-03" }, { "testId": "O6", "grade": "no", "says": "The Xiaomi MiMo platform privacy policy (effective 10 October 2026) says: \"Without your prior consent, Xiaomi will not use the text content provided by you for model training or any other purposes.\" The sentence covers text, not images, video or audio.", "source": "https://privacy.mi.com/XiaomiMiMoPlatform/en_GB/", "checked": "2026-11-03" }, { "testId": "O7", "grade": "yes", "says": "The card states MIT with no added license condition on derivatives; the use-restrictions notice lists prohibited uses but says it imposes no additional license conditions.", "source": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL", "checked": "2026-11-03" } ] }, { "slug": "phi-4-reasoning-vision", "product": "Phi-4-reasoning-vision-15B", "maker": "Microsoft", "what": "Microsoft's compact open-weight multimodal reasoning model: a 15B model combining the Phi-4-Reasoning language backbone with a SigLIP-2 image encoder, released 4 March 2026 under MIT.", "page": "https://thecontinuitylayer.com/open-models/phi-4-reasoning-vision", "docsUrl": "https://huggingface.co/microsoft/Phi-4-reasoning-vision-15B", "pricingUrl": "https://huggingface.co/microsoft/Phi-4-reasoning-vision-15B", "license": "MIT, stated on the model card.", "kind": null, "score": 6.5, "outOf": 7, "notStated": 0, "cells": [ { "testId": "O1", "grade": "yes", "says": "The card lists the license as MIT.", "source": "https://huggingface.co/microsoft/Phi-4-reasoning-vision-15B", "checked": "2026-11-03" }, { "testId": "O2", "grade": "yes", "says": "Weights are on Hugging Face under microsoft in BF16; the Hub lists eight quantized versions built from the model.", "source": "https://huggingface.co/microsoft/Phi-4-reasoning-vision-15B", "checked": "2026-11-03" }, { "testId": "O3", "grade": "yes", "says": "The card states a context length of 16,384 tokens.", "source": "https://huggingface.co/microsoft/Phi-4-reasoning-vision-15B", "checked": "2026-11-03" }, { "testId": "O4", "grade": "yes", "says": "15B parameters in BF16; the card says the model was tested on NVIDIA A6000, A100, H100 and B200 GPUs under Ubuntu 22.04.5 and that other GPUs with enough memory may work but are untested; it gives no memory figure or quantization guidance.", "source": "https://huggingface.co/microsoft/Phi-4-reasoning-vision-15B", "checked": "2026-11-03" }, { "testId": "O5", "grade": "partly", "says": "The card links the model on Azure AI Foundry and states no per-token price; no Microsoft page read on the day checked gives one.", "source": "https://huggingface.co/microsoft/Phi-4-reasoning-vision-15B", "checked": "2026-11-03" }, { "testId": "O6", "grade": "no", "says": "Microsoft's Azure AI Foundry data privacy page (last updated 2 March 2026) says of serverless API deployments: \"Microsoft doesn't use these prompts and outputs to train or improve Microsoft models, the model provider's models, or any third party's models.\"", "source": "https://learn.microsoft.com/azure/ai-foundry/how-to/concept-data-privacy", "checked": "2026-11-03" }, { "testId": "O7", "grade": "yes", "says": "MIT per the card, which states no condition on derivatives; the card gives no fine-tuning guidance of its own.", "source": "https://huggingface.co/microsoft/Phi-4-reasoning-vision-15B", "checked": "2026-11-03" } ] }, { "slug": "gemma-4", "product": "Gemma 4", "maker": "Google DeepMind", "what": "Google DeepMind's family of open models in five sizes (E2B, E4B, 12B, 26B A4B and 31B), with a 256K context window on the medium models and Apache 2.0 licensing; the two largest also run on the Gemini API.", "page": "https://thecontinuitylayer.com/open-models/gemma-4", "docsUrl": "https://huggingface.co/google/gemma-4-31B-it", "pricingUrl": "https://aistudio.google.com/docs/pricing", "license": "Apache 2.0; the Hugging Face card links Google's Gemma 4 license page, which carries the Apache License 2.0 text.", "kind": null, "score": 6, "outOf": 7, "notStated": 0, "cells": [ { "testId": "O1", "grade": "yes", "says": "The Hugging Face card lists the license as Apache 2.0 and links Google's Gemma 4 license page, which carries the Apache License 2.0 text with no added conditions; Google's Gemma docs say the models permit responsible commercial use.", "source": "https://ai.google.dev/gemma/docs/gemma_4_license", "checked": "2026-11-03" }, { "testId": "O2", "grade": "yes", "says": "Weights for all five sizes, base and instruction-tuned, are on Hugging Face under google in BF16, with a quantizations list for llama.cpp, Ollama and LM Studio; the card shows no access request step.", "source": "https://huggingface.co/google/gemma-4-31B-it", "checked": "2026-11-03" }, { "testId": "O3", "grade": "yes", "says": "The card says \"Gemma 4 features a context window of up to 256K tokens\": 256K on the 12B, 26B A4B and 31B models and 128K on E2B and E4B.", "source": "https://huggingface.co/google/gemma-4-31B-it", "checked": "2026-11-03" }, { "testId": "O4", "grade": "yes", "says": "Google's Gemma docs give the memory to load each size's weights: 31B needs 69.9GB in BF16, 34.9GB at 8-bit and 17.5GB at 4-bit; 26B A4B 57.7, 28.8 and 14.4GB; 12B 26.7, 13.4 and 6.7GB; E4B 17.9, 8.9 and 4.5GB; E2B 11.4, 5.7 and 2.9GB. The figures exclude the context window.", "source": "https://ai.google.dev/gemma/docs/core", "checked": "2026-11-03" }, { "testId": "O5", "grade": "partly", "says": "Google's Gemma docs say the Gemini API supports gemma-4-31b-it and gemma-4-26b-a4b-it with an API key from AI Studio; the page gives no price, and the Gemini API pricing page lists no Gemma model.", "source": "https://ai.google.dev/gemma/docs/core/gemma_on_gemini_api", "checked": "2026-11-03" }, { "testId": "O6", "grade": "partly", "says": "The Gemini API pricing page's tier overview says Free: \"Content used to improve our products\" and Paid: \"Content not used to improve our products\"; which tier a Gemma call falls under is not stated on the page.", "source": "https://aistudio.google.com/docs/pricing", "checked": "2026-11-03" }, { "testId": "O7", "grade": "yes", "says": "Apache 2.0 per the card, which states no condition on fine-tuning, derivatives or redistribution and lists downstream fine-tunes, adapters and merges in its model tree.", "source": "https://huggingface.co/google/gemma-4-26B-A4B-it", "checked": "2026-11-03" } ] }, { "slug": "deepseek-v4-1-flash", "product": "DeepSeek-V4.1-Flash", "maker": "DeepSeek", "what": "DeepSeek's current open release: a multimodal mixture-of-experts model with 552B backbone parameters, served on DeepSeek's API as deepseek-flash (version DeepSeek-V4.1-Flash-1017).", "page": "https://thecontinuitylayer.com/open-models/deepseek-v4-1-flash", "docsUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash", "pricingUrl": "https://api-docs.deepseek.com/quick_start/pricing.md", "license": "MIT License, stated on the model card for the repository and the weights.", "kind": null, "score": 5.5, "outOf": 7, "notStated": 1, "cells": [ { "testId": "O1", "grade": "yes", "says": "The card says the repository and the model weights are licensed under the MIT License, with the license file linked.", "source": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash", "checked": "2026-11-03" }, { "testId": "O2", "grade": "yes", "says": "Weights are on Hugging Face under deepseek-ai, with 8-bit and FP8 tags and an inference folder for local setup; DeepSeek-V4.1-Flash-1017 is a second repository.", "source": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash", "checked": "2026-11-03" }, { "testId": "O3", "grade": "yes", "says": "The card states support for contexts of up to one million tokens and lists a 1M-token context window in its sampling settings.", "source": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash", "checked": "2026-11-03" }, { "testId": "O4", "grade": "partly", "says": "552B backbone parameters, with 8B active per token during prefill and 16B during decode; the card gives no GPU memory figure and points to the inference folder and to community quantizations for llama.cpp, Ollama and LM Studio.", "source": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash", "checked": "2026-11-03" }, { "testId": "O5", "grade": "yes", "says": "DeepSeek's pricing page lists deepseek-flash (DeepSeek-V4.1-Flash-1017) at $0.30 per million input tokens on a cache miss and $1.20 per million output tokens at peak, half that off-peak, and $0.006 per million on a cache hit, with a 1M context and 384K max output.", "source": "https://api-docs.deepseek.com/quick_start/pricing.md", "checked": "2026-11-03" }, { "testId": "O6", "grade": "not stated", "says": "DeepSeek's Open Platform Terms of Service (effective 29 April 2026) do not say whether API inputs or outputs train DeepSeek's models and describe no opt-out; they defer personal information to the Privacy Policy, which says data is used to \"operate, provide, develop, and improve the Services.\"", "source": "https://cdn.deepseek.com/policies/en-US/deepseek-open-platform-terms-of-service.html", "checked": "2026-11-03" }, { "testId": "O7", "grade": "yes", "says": "The card places the repository and weights under the MIT License and states no condition on fine-tuning, derivatives or redistribution.", "source": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash", "checked": "2026-11-03" } ] }, { "slug": "mistral-large-4", "product": "Mistral Large 4", "maker": "Mistral AI", "what": "Mistral AI's 1-trillion-parameter natively multimodal mixture-of-experts model with about 52B active parameters and a 1M context, released on Hugging Face in BF16 and NVFP4; Mistral Small 4 (119B, Apache 2.0) is the smaller open release.", "page": "https://thecontinuitylayer.com/open-models/mistral-large-4", "docsUrl": "https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B", "pricingUrl": "https://docs.mistral.ai/inference/pricing", "license": "Mistral Ecosystem Open-Weight License, \"a modified version of the MIT license\"; Mistral Small 4 is Apache 2.0.", "kind": null, "score": 5, "outOf": 7, "notStated": 0, "cells": [ { "testId": "O1", "grade": "partly", "says": "Mistral Large 4 is under the Mistral Ecosystem Open-Weight License, which calls itself \"a modified version of the MIT license\" and requires a separate commercial license from Mistral before operating the model as a service for third parties. Mistral Small 4 is under Apache 2.0.", "source": "https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B/blob/main/LICENSE", "checked": "2026-11-03" }, { "testId": "O2", "grade": "yes", "says": "BF16 and NVFP4 weights are on Hugging Face under mistralai.", "source": "https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B", "checked": "2026-11-03" }, { "testId": "O3", "grade": "yes", "says": "The card states \"1M context length.\"", "source": "https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B", "checked": "2026-11-03" }, { "testId": "O4", "grade": "partly", "says": "About 1T total and 52B active parameters; the NVFP4 checkpoint targets Blackwell GPUs with a smaller memory footprint, the serving example uses tensor-parallel size 8, and the card gives no GPU memory requirement.", "source": "https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B", "checked": "2026-11-03" }, { "testId": "O5", "grade": "yes", "says": "Mistral's API pricing page lists Mistral Large 4 at $0.7 per million input tokens and $2.1 per million output tokens, cached input $0.07; Mistral Small 4 at $0.15 in and $0.6 out.", "source": "https://docs.mistral.ai/inference/pricing", "checked": "2026-11-03" }, { "testId": "O6", "grade": "partly", "says": "Mistral's help center says opting out of training for la Plateforme and the APIs is a toggle in the Admin Console's Privacy panel, that Teams and Enterprise are opted out by default, and that Free, Pro and Student plans must opt out themselves.", "source": "https://help.mistral.ai/en/articles/347614-can-i-opt-out-of-my-input-or-output-data-being-used-for-training", "checked": "2026-11-03" }, { "testId": "O7", "grade": "partly", "says": "The license grants the right to \"further train or otherwise modify the Model\", create derivative works and \"distribute, sublicense, and/or sell copies of the Model\" with the notice kept; offering the model or its derivatives as a service to third parties needs a separate commercial license.", "source": "https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B/blob/main/LICENSE", "checked": "2026-11-03" } ] }, { "slug": "gpt-oss-120b", "product": "gpt-oss-120b", "maker": "OpenAI", "what": "OpenAI's open-weight model for reasoning and agentic tasks: 117B total and 5.1B active parameters, released with the smaller gpt-oss-20b in August 2025 under Apache 2.0.", "page": "https://thecontinuitylayer.com/open-models/gpt-oss-120b", "docsUrl": "https://developers.openai.com/docs/models/gpt-oss", "pricingUrl": "https://huggingface.co/openai/gpt-oss-120b", "license": "Apache 2.0, stated on the model card and on OpenAI's model page.", "kind": null, "score": 5, "outOf": 7, "notStated": 1, "cells": [ { "testId": "O1", "grade": "yes", "says": "OpenAI's model page states a \"Permissive Apache 2.0 license\" and adds \"Build freely without copyleft restrictions or patent risk.\"", "source": "https://developers.openai.com/docs/models/gpt-oss", "checked": "2026-11-03" }, { "testId": "O2", "grade": "yes", "says": "Weights for gpt-oss-120b and gpt-oss-20b are on Hugging Face under openai, with MXFP4 quantization of the MoE weights.", "source": "https://huggingface.co/openai/gpt-oss-120b", "checked": "2026-11-03" }, { "testId": "O3", "grade": "yes", "says": "OpenAI's model page states a 131,072-token context window and 131,072 max output tokens.", "source": "https://developers.openai.com/api/docs/models/gpt-oss-120b.md", "checked": "2026-11-03" }, { "testId": "O4", "grade": "yes", "says": "117B total and 5.1B active parameters; the card says gpt-oss-120b is intended to fit into a single 80GB GPU such as an NVIDIA H100 or AMD MI300X, and gpt-oss-20b runs within 16GB of memory.", "source": "https://huggingface.co/openai/gpt-oss-120b", "checked": "2026-11-03" }, { "testId": "O5", "grade": "no", "says": "OpenAI's model page gives no per-token price for gpt-oss-120b and its rate-limit table shows 0 RPM and 0 TPM for every tier; the page links to Hugging Face for download.", "source": "https://developers.openai.com/api/docs/models/gpt-oss-120b.md", "checked": "2026-11-03" }, { "testId": "O6", "grade": "not stated", "says": "OpenAI does not serve gpt-oss on its API, so there is no hosted default to grade. OpenAI's API data privacy page (updated 8 January 2026) says of its API platform: \"We do not train our models on your data by default.\"", "source": "https://openai.com/api-data-privacy", "checked": "2026-11-03" }, { "testId": "O7", "grade": "yes", "says": "Apache 2.0 with no added condition; the card says \"Both gpt-oss models can be fine-tuned for a variety of specialized use cases\" and lists the models as fine-tunable.", "source": "https://huggingface.co/openai/gpt-oss-120b", "checked": "2026-11-03" } ] }, { "slug": "nemotron-3-5-lightning", "product": "NVIDIA Nemotron 3.5 Lightning", "maker": "NVIDIA", "what": "NVIDIA's 30B-total, 3B-active hybrid Mamba-2, mixture-of-experts and attention model for agents and local inference, released 11 August 2026 in BF16, NVFP4 and GGUF under the OpenMDW-1.1 license.", "page": "https://thecontinuitylayer.com/open-models/nemotron-3-5-lightning", "docsUrl": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16", "pricingUrl": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16", "license": "OpenMDW-1.1, the OpenMDW License Agreement version 1.1; the card says ready for commercial use; a permissive model license not on the OSI's approved list on the pages read.", "kind": null, "score": 5, "outOf": 7, "notStated": 1, "cells": [ { "testId": "O1", "grade": "partly", "says": "The card says use of the model is governed by the OpenMDW-1.1 license and that it is ready for commercial use. The license text grants permission to deal in the model materials without restriction, free of charge, provided the agreement and its notices stay with any distribution; it is a permissive license, but not an OSI-approved one on the pages read.", "source": "https://github.com/OpenMDW/openmdw/blob/main/1.1/LICENSE.OpenMDW-1.1", "checked": "2026-11-03" }, { "testId": "O2", "grade": "yes", "says": "BF16 reference weights, an NVFP4 release and a GGUF checkpoint are on Hugging Face under nvidia.", "source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16", "checked": "2026-11-03" }, { "testId": "O3", "grade": "yes", "says": "The card states \"Up to 1M tokens (for single H100 deployment, we use 256K).\"", "source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16", "checked": "2026-11-03" }, { "testId": "O4", "grade": "yes", "says": "30B total and 3B active parameters; the NVFP4 card lists a single H100 and A100-class GPUs, a Local AI section for RTX 5090, DGX Spark and RTX 6000 Pro, and Ollama defaults of 4K context under 24GB of VRAM, 32K at 24 to 48GB and 256K above 48GB.", "source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4", "checked": "2026-11-03" }, { "testId": "O5", "grade": "partly", "says": "The card links a chat demo on build.nvidia.com and states no per-token price; the build.nvidia.com page could not be opened on the day checked.", "source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16", "checked": "2026-11-03" }, { "testId": "O6", "grade": "not stated", "says": "No NVIDIA page read on the day checked says whether prompts sent to the build.nvidia.com endpoint train NVIDIA models.", "source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16", "checked": "2026-11-03" }, { "testId": "O7", "grade": "yes", "says": "The OpenMDW-1.1 text grants permission to deal in the model materials without restriction, with no condition on derivatives beyond keeping the agreement and notices in any distribution; the card lists post-training, domain adaptation and quantized or GGUF variants as intended uses.", "source": "https://github.com/OpenMDW/openmdw/blob/main/1.1/LICENSE.OpenMDW-1.1", "checked": "2026-11-03" } ] }, { "slug": "kimi-k3", "product": "Kimi K3", "maker": "Moonshot AI", "what": "Moonshot AI's open-weight native multimodal agentic model: 2.8T total and 104B active parameters with a 1-million-token context window, served on the Kimi Open Platform as kimi-k3.", "page": "https://thecontinuitylayer.com/open-models/kimi-k3", "docsUrl": "https://huggingface.co/moonshotai/Kimi-K3", "pricingUrl": "https://platform.kimi.ai/docs/pricing/chat-k3", "license": "Kimi K3 License, stated on the model card for both the code and the weights.", "kind": null, "score": 4.5, "outOf": 7, "notStated": 0, "cells": [ { "testId": "O1", "grade": "partly", "says": "Weights and code are under the Kimi K3 License, an MIT-style grant with conditions: a model-as-a-service business with over $20 million in revenue over 12 months needs a separate agreement with Moonshot AI, and a product with over 100 million monthly active users or $20 million monthly revenue must display \"Kimi K3\" on its interface.", "source": "https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE", "checked": "2026-11-03" }, { "testId": "O2", "grade": "yes", "says": "Weights are on Hugging Face under moonshotai, in MXFP4 from quantization-aware training.", "source": "https://huggingface.co/moonshotai/Kimi-K3", "checked": "2026-11-03" }, { "testId": "O3", "grade": "yes", "says": "The card states a 1-million-token context window and lists a context length of 1048576.", "source": "https://huggingface.co/moonshotai/Kimi-K3", "checked": "2026-11-03" }, { "testId": "O4", "grade": "partly", "says": "2.8T total and 104B active parameters in MXFP4 weights; the card gives no GPU memory figure and points to vLLM, SGLang and TokenSpeed recipes and to community quantizations for llama.cpp, Ollama and LM Studio.", "source": "https://huggingface.co/moonshotai/Kimi-K3", "checked": "2026-11-03" }, { "testId": "O5", "grade": "yes", "says": "Kimi's pricing article lists the kimi-k3 API at $3.00 per million input tokens on a cache miss, $0.30 on a cache hit, and $15.00 per million output tokens. The platform pricing page's table did not render as text on the day checked.", "source": "https://kimi.com/resources/kimi-k3-pricing", "checked": "2026-11-03" }, { "testId": "O6", "grade": "yes", "says": "The Kimi Open Platform model use agreement says Customer Content may be used to \"provide, maintain, develop, support, and improve the Services\" and that \"Unless otherwise expressly agreed in writing, Customer Content may be used for the foregoing purposes\"; a customer wanting restrictions on use for training must contact Moonshot AI.", "source": "https://platform.kimi.ai/docs/agreement/modeluse", "checked": "2026-11-03" }, { "testId": "O7", "grade": "partly", "says": "The license grants the right to \"run, deploy, fine-tune, or otherwise modify the Software and create derivative works\" and to furnish copies with the notice kept, subject to the revenue and user thresholds above.", "source": "https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE", "checked": "2026-11-03" } ] }, { "slug": "qwen3-8", "product": "Qwen3.8", "maker": "Alibaba (Qwen)", "what": "The open release of Alibaba's Qwen3.8 generation: a 2.4T-total, 95B-active flagship (Qwen3.8-2.4T-A95B) and a 27B native vision-language model (Qwen3.8-27B), both on Hugging Face, with Qwen Cloud as the managed API.", "page": "https://thecontinuitylayer.com/open-models/qwen3-8", "docsUrl": "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B", "pricingUrl": "https://www.qwencloud.com/models/qwen3.8-max", "license": "Qwen3.8-Max License; Apache-2.0 for Qwen3.8-27B.", "kind": null, "score": 4.5, "outOf": 7, "notStated": 1, "cells": [ { "testId": "O1", "grade": "partly", "says": "The 2.4T-A95B flagship is under the Qwen3.8-Max License, an MIT-style grant with conditions: a model-as-a-service or AI work assistant business with over $50 million in revenue over 12 months needs a separate license from Qwen, and a product with over 100 million monthly active users or $20 million monthly revenue must display the model name. Qwen3.8-27B is under Apache-2.0.", "source": "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B/blob/main/LICENSE", "checked": "2026-11-03" }, { "testId": "O2", "grade": "yes", "says": "Qwen3.8-2.4T-A95B, Qwen3.8-27B and FP8 versions of both are published on Hugging Face under Qwen, with a Qwen3.8-Flash-Next repository alongside.", "source": "https://huggingface.co/Qwen", "checked": "2026-11-03" }, { "testId": "O3", "grade": "yes", "says": "The card states \"Context Length: 262,144 natively and extensible up to 1,010,000 tokens.\" The 27B card states 262,144 natively, extensible to 1,000,000.", "source": "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B", "checked": "2026-11-03" }, { "testId": "O4", "grade": "partly", "says": "2.4T total and 95B active parameters for the flagship and 27B for the smaller model; the cards give no GPU memory figure, point to SGLang, vLLM and TokenSpeed for serving, and link community quantizations for llama.cpp, Ollama and LM Studio.", "source": "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B", "checked": "2026-11-03" }, { "testId": "O5", "grade": "yes", "says": "Qwen Cloud lists qwen3.8-max at $2 per million input tokens and $6 per million output tokens, with cached input at $0.25 and batch at $1 in and $3 out, on a 1M context. The 27B card says a hosted Qwen3.8-27B is coming soon.", "source": "https://www.qwencloud.com/models/qwen3.8-max", "checked": "2026-11-03" }, { "testId": "O6", "grade": "not stated", "says": "Qwen Cloud's data security page, the one its Trust Center links, covers encryption, regions, API keys and content moderation and does not say whether API inputs or outputs train its models; the privacy policy and terms pages did not render on the day checked. A copy of the same page on a second Qwen domain adds that they are not used for training.", "source": "https://docs.qwencloud.com/developer-guides/security-compliance/data-security", "checked": "2026-11-03" }, { "testId": "O7", "grade": "partly", "says": "The Qwen3.8-Max License permits modifying, fine-tuning, distributing, sublicensing and selling the software and derivative works, with the copyright and license notice kept in copies and the revenue and user thresholds above. Qwen3.8-27B under Apache-2.0 carries no threshold.", "source": "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B/blob/main/LICENSE", "checked": "2026-11-03" } ] }, { "slug": "glm-5-3", "product": "GLM-5.3", "maker": "Z.ai", "what": "Z.ai's current flagship open release, a 753B-parameter model post-trained from the same base as GLM-5.2; GLM-5.3-Flash is the 320B-total, 18B-active multimodal model in the same series.", "page": "https://thecontinuitylayer.com/open-models/glm-5-3", "docsUrl": "https://huggingface.co/zai-org/GLM-5.3", "pricingUrl": "https://docs.z.ai/guides/overview/pricing.md", "license": "GLM-5.3 License; MIT for GLM-5.3-Flash.", "kind": null, "score": 4, "outOf": 7, "notStated": 1, "cells": [ { "testId": "O1", "grade": "partly", "says": "GLM-5.3 is under the GLM-5.3 License, an MIT-style grant with one added condition: a model-as-a-service operator whose revenue exceeds 10 billion US dollars over 12 months must pass Z.AI's security review before commercial use. GLM-5.3-Flash is under MIT.", "source": "https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE", "checked": "2026-11-03" }, { "testId": "O2", "grade": "yes", "says": "GLM-5.3 and GLM-5.3-Flash weights are on Hugging Face under zai-org, each with a BF16 repository beside the FP8 one.", "source": "https://huggingface.co/zai-org", "checked": "2026-11-03" }, { "testId": "O3", "grade": "not stated", "says": "The card gives no context window for the model; it lists only evaluation settings of 300,000, 400K and 1M tokens, and the Z.ai pricing page gives no context length.", "source": "https://huggingface.co/zai-org/GLM-5.3", "checked": "2026-11-03" }, { "testId": "O4", "grade": "partly", "says": "753B total parameters with no active count on the card and no GPU memory figure; deployment is listed for SGLang, vLLM, TokenSpeed, Transformers, KTransformers, Unsloth and Ascend NPU, with community quantizations for llama.cpp, Ollama and LM Studio.", "source": "https://huggingface.co/zai-org/GLM-5.3", "checked": "2026-11-03" }, { "testId": "O5", "grade": "yes", "says": "Z.ai's pricing page lists GLM-5.3 at $1.40 per million input tokens and $4.40 per million output tokens, cached input $0.26; GLM-5.3-Flash at $0.15 in and $0.50 out.", "source": "https://docs.z.ai/guides/overview/pricing.md", "checked": "2026-11-03" }, { "testId": "O6", "grade": "partly", "says": "Z.ai's privacy policy (last updated 29 September 2025) lists training among its improvement purposes for individual users; for API customers the Data Processing Addendum says \"The Company do not store any of the content the Customer or its End Users provide\" and does not mention training.", "source": "https://docs.z.ai/legal-agreement/privacy-policy.md", "checked": "2026-11-03" }, { "testId": "O7", "grade": "partly", "says": "The license grants the right \"to run, deploy, fine-tune, or otherwise modify the Software and create derivative works from it\" and to copy, distribute, sublicense and sell, with the notice kept in copies; the 10 billion dollar model-as-a-service clause is the only threshold.", "source": "https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE", "checked": "2026-11-03" } ] }, { "slug": "muse-spark-open", "product": "Muse Spark Open", "maker": "Meta", "what": "Meta's open-weight model from Meta Superintelligence Labs, released 15 October 2026: 816B total and 42B active parameters with a 1M-token context window. It is Meta's newer open release than the Llama 4 models of April 2025.", "page": "https://thecontinuitylayer.com/open-models/muse-spark-open", "docsUrl": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B", "pricingUrl": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B/blob/main/LICENSE.md", "license": "Muse Model License, version Muse Spark Open, release date 15 October 2026.", "kind": null, "score": 3.5, "outOf": 7, "notStated": 1, "cells": [ { "testId": "O1", "grade": "partly", "says": "The weights are under the Muse Model License, a custom license: it grants use, modification, fine-tuning and distribution, requires \"Built with Muse Spark Open\" in the user interface, incorporates a Usage Policy, and asks a model-as-a-service business with over $20 million in gross revenue over 12 months to sign a separate agreement with Meta.", "source": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B/blob/main/LICENSE.md", "checked": "2026-11-03" }, { "testId": "O2", "grade": "partly", "says": "BF16 and NVFP4 weights are on Hugging Face under meta-models, gated: you log in, accept the conditions and submit a form with your legal name, date of birth and organization before access.", "source": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B", "checked": "2026-11-03" }, { "testId": "O3", "grade": "yes", "says": "The card states a context window of 1M tokens, \"1M at inference.\"", "source": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B", "checked": "2026-11-03" }, { "testId": "O4", "grade": "yes", "says": "816B total and 42B active parameters; the card says NVFP4 needs a GPU with at least 700GB of VRAM and BF16 needs 2TB, and that the model is optimized to run on two Mac Studios with M5 Ultra using MLX.", "source": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B", "checked": "2026-11-03" }, { "testId": "O5", "grade": "no", "says": "Meta's Model API page says of Muse Spark Open: \"You don't call it over Model API\"; you self-host it or use a hosted inference provider. Meta's per-token prices cover the closed Muse Spark 1.1, 1.2 and 1.3.", "source": "https://dev.meta.ai/docs/models", "checked": "2026-11-03" }, { "testId": "O6", "grade": "not stated", "says": "Meta does not host this model, so there is no hosted default to grade. For the closed Muse Spark on Meta's Model API, Standard mode does not use your content to train Meta models and Contributor mode may, at a lower price.", "source": "https://dev.meta.ai/docs/models", "checked": "2026-11-03" }, { "testId": "O7", "grade": "partly", "says": "The license grants \"the rights to use, copy, modify (including fine-tune), create derivative works, distill, merge, publish, distribute, sublicense, and/or sell copies of the Software\", with conditions: the \"Built with Muse Spark Open\" display, the Usage Policy, and the $20 million model-as-a-service clause.", "source": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B/blob/main/LICENSE.md", "checked": "2026-11-03" } ] }, { "slug": "olmo-3-1", "product": "Olmo 3.1 32B", "maker": "Ai2 (Allen Institute for AI)", "what": "Ai2's fully open 32B language model, released with its training data (Dolma 3, Dolci), code and checkpoints under Apache 2.0, in Instruct and Think variants alongside Olmo 3 7B.", "page": "https://thecontinuitylayer.com/open-models/olmo-3-1", "docsUrl": "https://huggingface.co/allenai/Olmo-3.1-32B-Instruct", "pricingUrl": "https://huggingface.co/allenai/Olmo-3.1-32B-Instruct", "license": "Apache 2.0, stated on the model card, with Ai2's Responsible Use Guidelines linked.", "kind": null, "score": 3.5, "outOf": 7, "notStated": 2, "cells": [ { "testId": "O1", "grade": "yes", "says": "The card lists the license as Apache 2.0 and links Ai2's Responsible Use Guidelines.", "source": "https://huggingface.co/allenai/Olmo-3.1-32B-Instruct", "checked": "2026-11-03" }, { "testId": "O2", "grade": "yes", "says": "Weights are on Hugging Face under allenai in BF16, with the Dolma 3 pretraining and Dolci post-training datasets and the training code linked from the card.", "source": "https://huggingface.co/allenai/Olmo-3.1-32B-Instruct", "checked": "2026-11-03" }, { "testId": "O3", "grade": "not stated", "says": "The card gives no context window.", "source": "https://huggingface.co/allenai/Olmo-3.1-32B-Instruct", "checked": "2026-11-03" }, { "testId": "O4", "grade": "partly", "says": "32B parameters in BF16; the card shows 8-bit loading with bitsandbytes and gives no GPU memory figure or hardware list.", "source": "https://huggingface.co/allenai/Olmo-3.1-32B-Instruct", "checked": "2026-11-03" }, { "testId": "O5", "grade": "no", "says": "The card lists no Ai2-hosted API or price and says the model is not deployed by any inference provider on Hugging Face.", "source": "https://huggingface.co/allenai/Olmo-3.1-32B-Instruct", "checked": "2026-11-03" }, { "testId": "O6", "grade": "not stated", "says": "Ai2 hosts no API for the model on the pages read, so there is no hosted default to grade.", "source": "https://huggingface.co/allenai/Olmo-3.1-32B-Instruct", "checked": "2026-11-03" }, { "testId": "O7", "grade": "yes", "says": "Apache 2.0 per the card, which states no condition on derivatives beyond the linked Responsible Use Guidelines.", "source": "https://huggingface.co/allenai/Olmo-3.1-32B-Instruct", "checked": "2026-11-03" } ] } ] }, { "slug": "inference-apis", "name": "Inference APIs", "page": "https://thecontinuitylayer.com/inference-apis", "checked": "2026-11-03", "tests": [ { "id": "I1", "short": "Price", "question": "Does {name} publish a per-token price?", "yes": "Per-token prices are on a public page with no sign-in. Partly means only some models are priced in public, or the price sits on a model page rather than the pricing page." }, { "id": "I2", "short": "Rate limits", "question": "Does {name} publish its rate limits?", "yes": "The docs give request or token numbers for each tier or model. Partly means some numbers, such as a concurrency cap or the top tiers only; no means the numbers are only shown inside a signed-in account." }, { "id": "I3", "short": "Training", "question": "Does {name} train on your API prompts by default?", "yes": "Here no is the good answer: the provider's page says what you send its API is not used for training unless you opt in. Yes means it is unless you opt out; partly means it depends on the plan, the tier or the upstream provider.", "invert": true }, { "id": "I4", "short": "Retention", "question": "Does {name} say how long it keeps your prompts and outputs?", "yes": "A stated period, or a statement that nothing is kept by default. Partly means a period for some users or requests but not others." }, { "id": "I5", "short": "Status page", "question": "Does {name} run a public status page?", "yes": "A public page shows current uptime and past incidents." } ], "entries": [ { "slug": "groq", "product": "Groq", "maker": "Groq", "kind": "Inference host", "what": "An inference API that serves open models on Groq's own hardware, billed per token.", "page": "https://thecontinuitylayer.com/inference-apis/groq", "docsUrl": "https://groq.com", "pricingUrl": "https://console.groq.com/docs/models", "license": null, "score": 5, "outOf": 5, "notStated": 0, "cells": [ { "testId": "I1", "grade": "yes", "says": "The public models page lists rates per 1M tokens; GPT OSS 120B costs $0.15 input and $0.60 output, though some enterprise models say contact sales.", "source": "https://console.groq.com/docs/models", "checked": "2026-11-03" }, { "testId": "I2", "grade": "yes", "says": "The docs publish request and token limits for each model, for example 30 RPM and 8K TPM for gpt-oss-120b, with a note that exceptions may apply.", "source": "https://console.groq.com/docs/rate-limits", "checked": "2026-11-03" }, { "testId": "I3", "grade": "no", "says": "The services agreement bars Groq from using customer inputs or outputs to train any model unless the customer gives explicit permission.", "source": "https://console.groq.com/docs/legal/services-agreement", "checked": "2026-11-03" }, { "testId": "I4", "grade": "yes", "says": "Inference data is not retained by default, but inputs and outputs may be logged for up to 30 days for troubleshooting or abuse checks; every customer can switch on zero retention.", "source": "https://console.groq.com/docs/your-data", "checked": "2026-11-03" }, { "testId": "I5", "grade": "yes", "says": "A public status page shows uptime for each production and preview model.", "source": "https://groqstatus.com", "checked": "2026-11-03" } ] }, { "slug": "openai-api", "product": "OpenAI API", "maker": "OpenAI", "kind": "Maker's API", "what": "OpenAI's own API for its GPT models.", "page": "https://thecontinuitylayer.com/inference-apis/openai-api", "docsUrl": "https://openai.com/api/", "pricingUrl": "https://developers.openai.com/api/docs/pricing", "license": null, "score": 5, "outOf": 5, "notStated": 0, "cells": [ { "testId": "I1", "grade": "yes", "says": "A public pricing page lists per-token rates with no sign-in, e.g. gpt-6.1-sol at $2.00 input / $10.00 output per 1M tokens.", "source": "https://developers.openai.com/api/docs/pricing", "checked": "2026-11-03" }, { "testId": "I2", "grade": "yes", "says": "The docs publish requests-per-minute and tokens-per-minute figures for each paid usage tier, with exact per-model limits shown in the account settings.", "source": "https://developers.openai.com/api/docs/guides/rate-limits", "checked": "2026-11-03" }, { "testId": "I3", "grade": "no", "says": "OpenAI says API data is not used to train its models unless the customer opts in.", "source": "https://developers.openai.com/api/docs/guides/your-data", "checked": "2026-11-03" }, { "testId": "I4", "grade": "yes", "says": "Abuse-monitoring logs are kept for up to 30 days by default, and zero data retention is offered to approved customers.", "source": "https://developers.openai.com/api/docs/guides/your-data", "checked": "2026-11-03" }, { "testId": "I5", "grade": "yes", "says": "OpenAI runs a public status page with a separate group for API components.", "source": "https://status.openai.com/", "checked": "2026-11-03" } ] }, { "slug": "anthropic-api", "product": "Anthropic API", "maker": "Anthropic", "kind": "Maker's API", "what": "Anthropic's own API for its Claude models.", "page": "https://thecontinuitylayer.com/inference-apis/anthropic-api", "docsUrl": "https://platform.claude.com/", "pricingUrl": "https://platform.claude.com/docs/en/about-claude/pricing", "license": null, "score": 5, "outOf": 5, "notStated": 0, "cells": [ { "testId": "I1", "grade": "yes", "says": "The public docs list per-token prices for every model, e.g. Claude Sonnet 5.5 at $2 input / $10 output per 1M tokens.", "source": "https://platform.claude.com/docs/en/about-claude/pricing", "checked": "2026-11-03" }, { "testId": "I2", "grade": "yes", "says": "The docs publish per-model limits on requests, input tokens and output tokens for each usage tier.", "source": "https://platform.claude.com/docs/en/api/rate-limits", "checked": "2026-11-03" }, { "testId": "I3", "grade": "no", "says": "Anthropic says it does not train on API inputs or outputs unless the customer opts in or sends feedback.", "source": "https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training", "checked": "2026-11-03" }, { "testId": "I4", "grade": "yes", "says": "Anthropic states retention for API data, with deletion within 30 days by default on its privacy pages and zero retention available by agreement; its API docs describe no retention by default except for some models that require 30 days.", "source": "https://platform.claude.com/docs/en/manage-claude/api-and-data-retention", "checked": "2026-11-03" }, { "testId": "I5", "grade": "yes", "says": "Anthropic runs a public status page that lists the Claude API as its own component.", "source": "https://status.claude.com/", "checked": "2026-11-03" } ] }, { "slug": "fireworks-ai", "product": "Fireworks AI", "maker": "Fireworks AI", "kind": "Inference host", "what": "A hosted inference platform for open models, with serverless per-token pricing and dedicated deployments.", "page": "https://thecontinuitylayer.com/inference-apis/fireworks-ai", "docsUrl": "https://fireworks.ai", "pricingUrl": "https://docs.fireworks.ai/serverless/pricing", "license": null, "score": 4.5, "outOf": 5, "notStated": 0, "cells": [ { "testId": "I1", "grade": "yes", "says": "A public docs page lists serverless rates per model; OpenAI GPT OSS 120B costs $0.15 per 1M input tokens and $0.60 per 1M output tokens in standard mode.", "source": "https://docs.fireworks.ai/serverless/pricing", "checked": "2026-11-03" }, { "testId": "I2", "grade": "partly", "says": "The docs publish request caps (10 RPM with no payment method, 6,000 RPM with credits) and token ceilings for Tier 3 and up, but give no numbers for lower tiers.", "source": "https://docs.fireworks.ai/serverless/rate-limits", "checked": "2026-11-03" }, { "testId": "I3", "grade": "no", "says": "The privacy notice says prompts and API inputs are not used to train its models without explicit opt-in, and the inference product page says it never trains on customer data.", "source": "https://fireworks.ai/privacy-policy", "checked": "2026-11-03" }, { "testId": "I4", "grade": "yes", "says": "Prompts and outputs for open models stay in memory only for the request, not logged or stored without opt-in; Responses API data saved with store=true is kept 30 days.", "source": "https://docs.fireworks.ai/guides/security_compliance/data_handling", "checked": "2026-11-03" }, { "testId": "I5", "grade": "yes", "says": "A public status page shows uptime for each serverless model and lists active incidents.", "source": "https://status.fireworks.ai", "checked": "2026-11-03" } ] }, { "slug": "openrouter", "product": "OpenRouter", "maker": "OpenRouter", "kind": "Router", "what": "One API key that routes requests to many model providers, with the per-token price set by the provider used.", "page": "https://thecontinuitylayer.com/inference-apis/openrouter", "docsUrl": "https://openrouter.ai", "pricingUrl": "https://openrouter.ai/openai/gpt-oss-120b", "license": null, "score": 4.5, "outOf": 5, "notStated": 0, "cells": [ { "testId": "I1", "grade": "yes", "says": "Each public model page lists rates per 1M tokens for every upstream provider; gpt-oss-120b's headline rate is $0.027 input and $0.098 output, and the rate depends on the provider used.", "source": "https://openrouter.ai/openai/gpt-oss-120b", "checked": "2026-11-03" }, { "testId": "I2", "grade": "yes", "says": "The docs publish free-model limits (20 requests a minute; 50 a day, or 1,000 after buying 10 credits) and say paid models have no platform-level request cap.", "source": "https://openrouter.ai/docs/api_reference/limits", "checked": "2026-11-03" }, { "testId": "I3", "grade": "partly", "says": "OpenRouter itself does not train on prompts unless you opt in, but some upstream providers it routes to may keep data for training, which account settings control.", "source": "https://openrouter.ai/docs/guides/privacy/data-collection", "checked": "2026-11-03" }, { "testId": "I4", "grade": "yes", "says": "Prompts and responses are not stored by default; if you opt in to logging, they are kept at least three months.", "source": "https://openrouter.ai/docs/guides/features/input-output-logging", "checked": "2026-11-03" }, { "testId": "I5", "grade": "yes", "says": "A public status page shows real-time and past system health.", "source": "https://status.openrouter.ai", "checked": "2026-11-03" } ] }, { "slug": "deepinfra", "product": "DeepInfra", "maker": "DeepInfra", "kind": "Inference host", "what": "A hosted inference platform for open models, billed per token.", "page": "https://thecontinuitylayer.com/inference-apis/deepinfra", "docsUrl": "https://deepinfra.com", "pricingUrl": "https://deepinfra.com/pricing", "license": null, "score": 4.5, "outOf": 5, "notStated": 0, "cells": [ { "testId": "I1", "grade": "yes", "says": "The public pricing page lists rates per model; Llama-3.3-70B-Instruct-Turbo costs $0.10 per 1M input tokens and $0.32 per 1M output tokens.", "source": "https://deepinfra.com/pricing", "checked": "2026-11-03" }, { "testId": "I2", "grade": "yes", "says": "The docs publish a default limit of 200 requests at once per model, and say how to ask for more.", "source": "https://docs.deepinfra.com/account/rate-limits", "checked": "2026-11-03" }, { "testId": "I3", "grade": "partly", "says": "DeepInfra says it does not train on data sent to its APIs, except Google and Anthropic models, where those companies' training policies apply.", "source": "https://docs.deepinfra.com/account/data-privacy", "checked": "2026-11-03" }, { "testId": "I4", "grade": "yes", "says": "Standard requests are held only in memory and deleted after the response, while image outputs and bulk jobs are kept for a short period it does not define.", "source": "https://docs.deepinfra.com/account/data-privacy", "checked": "2026-11-03" }, { "testId": "I5", "grade": "yes", "says": "A public status page shows API and website uptime plus health checks for each model.", "source": "https://status.deepinfra.com", "checked": "2026-11-03" } ] }, { "slug": "gemini-api", "product": "Gemini API", "maker": "Google", "kind": "Maker's API", "what": "Google's own API for its Gemini models, served through AI Studio.", "page": "https://thecontinuitylayer.com/inference-apis/gemini-api", "docsUrl": "https://ai.google.dev/", "pricingUrl": "https://ai.google.dev/gemini-api/docs/pricing", "license": null, "score": 4, "outOf": 5, "notStated": 0, "cells": [ { "testId": "I1", "grade": "yes", "says": "A public pricing page lists per-token rates, e.g. Gemini 4 Argon at $2.00 input / $10.00 output per 1M tokens through November 30, 2026, doubling December 1.", "source": "https://ai.google.dev/gemini-api/docs/pricing", "checked": "2026-11-03" }, { "testId": "I2", "grade": "partly", "says": "The docs publish tier rules, spend caps and batch limits, but send developers to AI Studio for the per-model request and token numbers.", "source": "https://ai.google.dev/gemini-api/docs/rate-limits", "checked": "2026-11-03" }, { "testId": "I3", "grade": "partly", "says": "Google says paid-tier prompts are not used to improve its products, but free-tier content is.", "source": "https://ai.google.dev/gemini-api/terms", "checked": "2026-11-03" }, { "testId": "I4", "grade": "yes", "says": "Google states that prompts and outputs are kept for 55 days for abuse monitoring, and 30 days when Search or Maps grounding is used.", "source": "https://ai.google.dev/gemini-api/docs/usage-policies", "checked": "2026-11-03" }, { "testId": "I5", "grade": "yes", "says": "Google's own troubleshooting docs link to a public AI Studio status page.", "source": "https://aistudio.google.com/status", "checked": "2026-11-03" } ] }, { "slug": "xai-api", "product": "xAI API", "maker": "xAI", "kind": "Maker's API", "what": "xAI's own API for its Grok models.", "page": "https://thecontinuitylayer.com/inference-apis/xai-api", "docsUrl": "https://x.ai/api", "pricingUrl": "https://docs.x.ai/developers/pricing", "license": null, "score": 4, "outOf": 5, "notStated": 1, "cells": [ { "testId": "I1", "grade": "yes", "says": "Public docs list per-token prices, e.g. grok-4.7 at $2.00 input / $6.00 output per 1M tokens for prompts under 200k tokens.", "source": "https://docs.x.ai/developers/pricing", "checked": "2026-11-03" }, { "testId": "I2", "grade": "yes", "says": "The docs publish per-model requests-per-second and tokens-per-minute limits for each spend tier.", "source": "https://docs.x.ai/docs/key-information/consumption-and-rate-limits", "checked": "2026-11-03" }, { "testId": "I3", "grade": "no", "says": "xAI says it never trains on API inputs or outputs without the customer's explicit permission.", "source": "https://docs.x.ai/developers/faq/security", "checked": "2026-11-03" }, { "testId": "I4", "grade": "yes", "says": "Requests and responses are stored encrypted for 30 days for abuse auditing, then deleted, and a team-wide zero-retention setting is available.", "source": "https://docs.x.ai/developers/faq/security", "checked": "2026-11-03" }, { "testId": "I5", "grade": "not stated", "says": "The status page would not open when checked.", "source": "https://status.x.ai/", "checked": "2026-11-03" } ] }, { "slug": "together-ai", "product": "Together AI", "maker": "Together AI", "kind": "Inference host", "what": "A hosted inference platform that serves open models through one API, billed per token.", "page": "https://thecontinuitylayer.com/inference-apis/together-ai", "docsUrl": "https://www.together.ai", "pricingUrl": "https://www.together.ai/pricing", "license": null, "score": 3.5, "outOf": 5, "notStated": 0, "cells": [ { "testId": "I1", "grade": "yes", "says": "The public pricing page lists serverless rates per model without a sign-in; gpt-oss-120B costs $0.15 per 1M input tokens and $0.60 per 1M output tokens.", "source": "https://www.together.ai/pricing", "checked": "2026-11-03" }, { "testId": "I2", "grade": "no", "says": "The rate-limits doc publishes no request or token limits per tier; it calls serverless use best-effort and says busy periods can return 429 or 503 errors.", "source": "https://docs.together.ai/docs/rate-limits", "checked": "2026-11-03" }, { "testId": "I3", "grade": "no", "says": "Sharing data to train models is opt-in and off by default, though stored prompts may still be used for product improvements.", "source": "https://docs.together.ai/docs/privacy-and-security", "checked": "2026-11-03" }, { "testId": "I4", "grade": "partly", "says": "Prompts and responses are stored by default with no stated time limit; customers can delete them or turn on zero data retention.", "source": "https://docs.together.ai/docs/zero-data-retention", "checked": "2026-11-03" }, { "testId": "I5", "grade": "yes", "says": "A public status page shows component uptime, incidents and subscription options.", "source": "https://status.together.ai", "checked": "2026-11-03" } ] }, { "slug": "mistral-api", "product": "Mistral API", "maker": "Mistral AI", "kind": "Maker's API", "what": "Mistral AI's own API platform, La Plateforme, for its models.", "page": "https://thecontinuitylayer.com/inference-apis/mistral-api", "docsUrl": "https://mistral.ai/", "pricingUrl": "https://docs.mistral.ai/inference/pricing", "license": null, "score": 3.5, "outOf": 5, "notStated": 0, "cells": [ { "testId": "I1", "grade": "yes", "says": "Public docs list per-token prices, e.g. Mistral Large 4 at $0.70 input / $2.10 output per 1M tokens.", "source": "https://docs.mistral.ai/inference/pricing", "checked": "2026-11-03" }, { "testId": "I2", "grade": "no", "says": "The docs name the kinds of limits that apply but give no numbers, saying the values are only shown in the signed-in admin panel.", "source": "https://docs.mistral.ai/admin/billing-usage/usage-limits", "checked": "2026-11-03" }, { "testId": "I3", "grade": "partly", "says": "The privacy policy says inputs and outputs may be used for training unless the customer opts out in account settings.", "source": "https://legal.mistral.ai/terms/privacy-policy/", "checked": "2026-11-03" }, { "testId": "I4", "grade": "yes", "says": "For most APIs, inputs and outputs are kept for 30 rolling days for abuse monitoring unless zero data retention is approved.", "source": "https://legal.mistral.ai/terms/privacy-policy/", "checked": "2026-11-03" }, { "testId": "I5", "grade": "yes", "says": "Mistral runs a public status page that lists API components such as chat completions, embeddings and batch.", "source": "https://status.mistral.ai/", "checked": "2026-11-03" } ] }, { "slug": "cohere-api", "product": "Cohere API", "maker": "Cohere", "kind": "Maker's API", "what": "Cohere's own API for its Command models.", "page": "https://thecontinuitylayer.com/inference-apis/cohere-api", "docsUrl": "https://cohere.com/", "pricingUrl": "https://docs.cohere.com/docs/command-a", "license": null, "score": 3.5, "outOf": 5, "notStated": 0, "cells": [ { "testId": "I1", "grade": "partly", "says": "The pricing page shows only legacy models, but a docs model page gives Command A at $2.50 input / $10 output per 1M tokens.", "source": "https://docs.cohere.com/docs/command-a", "checked": "2026-11-03" }, { "testId": "I2", "grade": "yes", "says": "The docs publish per-endpoint and per-model limits for trial and production keys, though some newer models say to contact sales.", "source": "https://docs.cohere.com/docs/rate-limits", "checked": "2026-11-03" }, { "testId": "I3", "grade": "partly", "says": "Paying customers can opt out of training in dashboard settings, while trial-user data may be used for research.", "source": "https://cohere.com/enterprise-data-commitments", "checked": "2026-11-03" }, { "testId": "I4", "grade": "partly", "says": "Paying customers' logged prompts and outputs are deleted after 30 days, but no period is given for trial users.", "source": "https://cohere.com/enterprise-data-commitments", "checked": "2026-11-03" }, { "testId": "I5", "grade": "yes", "says": "Cohere runs a public status page that lists its API endpoints and models.", "source": "https://status.cohere.com/", "checked": "2026-11-03" } ] }, { "slug": "baseten", "product": "Baseten", "maker": "Baseten", "kind": "Inference host", "what": "An inference platform for running models in production, with a per-token Model API and dedicated deployments.", "page": "https://thecontinuitylayer.com/inference-apis/baseten", "docsUrl": "https://www.baseten.co", "pricingUrl": "https://www.baseten.co/resources/changelog/elephant-1-and-mammoth-1-pricing/", "license": null, "score": 3.5, "outOf": 5, "notStated": 1, "cells": [ { "testId": "I1", "grade": "partly", "says": "The main pricing page would not open without JavaScript, but an official changelog lists elephant-1 at $0.035 per 1M input tokens; it gives no output price because the model returns decisions, not text.", "source": "https://www.baseten.co/resources/changelog/elephant-1-and-mammoth-1-pricing/", "checked": "2026-11-03" }, { "testId": "I2", "grade": "yes", "says": "The docs publish default limits for each account tier, from 15 RPM and 100K TPM unverified up to 120 RPM and 1M TPM on Pro.", "source": "https://docs.baseten.co/inference/model-apis/pricing-and-limits", "checked": "2026-11-03" }, { "testId": "I3", "grade": "not stated", "says": "None of the official pages that would open said whether Baseten trains on customer prompts or outputs; the terms, privacy and security-practices pages would not open.", "source": "https://docs.baseten.co/observability/security", "checked": "2026-11-03" }, { "testId": "I4", "grade": "yes", "says": "Baseten says it does not store inputs or outputs for synchronous Model API requests, under a zero data retention policy.", "source": "https://docs.baseten.co/inference/model-apis/overview", "checked": "2026-11-03" }, { "testId": "I5", "grade": "yes", "says": "A public status page shows Model APIs, inference and regional uptime, plus past incidents.", "source": "https://status.baseten.co", "checked": "2026-11-03" } ] }, { "slug": "deepseek-api", "product": "DeepSeek API", "maker": "DeepSeek", "kind": "Maker's API", "what": "DeepSeek's own API for its models.", "page": "https://thecontinuitylayer.com/inference-apis/deepseek-api", "docsUrl": "https://api-docs.deepseek.com/", "pricingUrl": "https://api-docs.deepseek.com/quick_start/pricing/", "license": null, "score": 2.5, "outOf": 5, "notStated": 2, "cells": [ { "testId": "I1", "grade": "yes", "says": "The public docs list per-token prices, e.g. deepseek-v4-pro at $1.32 input (cache miss) / $3.96 output per 1M tokens at peak, half that off-peak.", "source": "https://api-docs.deepseek.com/quick_start/pricing/", "checked": "2026-11-03" }, { "testId": "I2", "grade": "partly", "says": "The docs publish only concurrency caps per model, with no request-per-minute or token limits.", "source": "https://api-docs.deepseek.com/quick_start/rate_limit/", "checked": "2026-11-03" }, { "testId": "I3", "grade": "not stated", "says": "The API terms say nothing about training on customer data, and the separate Open Platform privacy policy page would not open.", "source": "https://cdn.deepseek.com/policies/en-US/deepseek-open-platform-terms-of-service.html", "checked": "2026-11-03" }, { "testId": "I4", "grade": "not stated", "says": "The API terms give no retention period for prompts or outputs, and the Open Platform privacy policy page would not open.", "source": "https://cdn.deepseek.com/policies/en-US/deepseek-open-platform-terms-of-service.html", "checked": "2026-11-03" }, { "testId": "I5", "grade": "yes", "says": "DeepSeek runs a public status page that lists its API services.", "source": "https://status.deepseek.com/", "checked": "2026-11-03" } ] }, { "slug": "hiapi", "product": "HiAPI", "maker": "HiAPI", "kind": "Router", "what": "One API key for text, image, video and audio models from many providers, with text models billed per token and media models per image, per second or per request.", "page": "https://thecontinuitylayer.com/inference-apis/hiapi", "docsUrl": "https://hiapi.ai/docs", "pricingUrl": "https://hiapi.ai/en/pricing", "license": null, "score": 2, "outOf": 5, "notStated": 1, "cells": [ { "testId": "I1", "grade": "yes", "says": "The public pricing page lists text-model rates per 1M tokens without a sign-in; deepseek-v4-flash costs $0.14 input and $0.28 output, while image, video and audio models are priced per image, per second or per request.", "source": "https://hiapi.ai/en/pricing", "checked": "2026-11-05" }, { "testId": "I2", "grade": "partly", "says": "The docs publish one account-wide limit of 20 write and 60 read requests per 10 seconds for the task and storage endpoints, with no token limits and no numbers per tier or model.", "source": "https://hiapi.ai/docs/rate-limits", "checked": "2026-11-05" }, { "testId": "I3", "grade": "not stated", "says": "Neither the privacy policy nor the terms say whether prompts or outputs sent through the API are used for training, by HiAPI or by the upstream providers whose models it serves.", "source": "https://hiapi.ai/en/privacy", "checked": "2026-11-05" }, { "testId": "I4", "grade": "partly", "says": "The privacy policy keeps API call logs for 90 days by default without saying whether those logs hold the prompt text, and the storage docs expire temporary outputs after about 7 days unless you pay to keep them; no period is stated for prompts themselves.", "source": "https://hiapi.ai/en/privacy", "checked": "2026-11-05" }, { "testId": "I5", "grade": "no", "says": "No public status page was found: the site links none, and status.hiapi.ai is a sign-in page for HiAPI's own team with no uptime or incident history.", "source": "https://status.hiapi.ai", "checked": "2026-11-05" } ] } ] } ] }