The Continuity Layer
The Scorecard · Coding agents

What your coding agent remembers between sessions, and who owns it

thecontinuitylayer.com/coding-agents

Claude Code, Cursor, Codex, Copilot and four more, read from their own docs and graded on the same eight questions: does it carry memory into the next session, can you see, edit and delete it, is it a file you own, does another tool read it, is your code used for training, and what it costs.

8
Coding agents graded
8
Tests
64
Sourced answers
Docs
Basis
Read from the makers' own pages on 3 November 2026

Every agent here can carry instructions into the next session, and most keep them in a plain file you own; the differences are in whether auto-generated memory can be seen and deleted, and whether your code trains the maker's models by default. The grades are read from each maker's own pages on the day shown, never from a review.

At a glance

Every product, every test.

✓ yes, ½ partly, ✕ no, – the maker's pages do not say. A dash counts for nothing: a product gets credit only for what its own pages state, so a vague page never outranks a clear one. On the training test the colors flip, because no is the good answer there.

Sort by
Coding agents: each product's grade on each test, and its score out of all the tests
ProductRemembersSee itEdit itDelete itYour filePortableTrainingPriceScore
Claude Code#1 · AnthropicA7½/8
OpenAI Codex#2 · OpenAIA−7/8
Cline#3 · Cline Bot Inc.A−7/8
Windsurf (now Devin Desktop)#4 · CognitionB+6½/8
Gemini CLI#5 · GoogleB+6½/8
Aider#6 · Aider-AIB+6½/8
Cursor#7 · AnysphereB6/8
GitHub Copilot#8 · GitHub (Microsoft)B6/8
Product by product

In score order, best first.

  • Claude Code

    #1 · Anthropic

    Anthropic's agentic coding tool: reads your codebase, edits files and runs commands from the terminal, an IDE, the desktop app or the browser.

    A7½/8On the 8 tests
    • Does Claude Code remember across sessions?Yes
    • Can you see what Claude Code remembered?Yes
    • Can you edit Claude Code's memory?Yes
    All 8 answers →
  • OpenAI Codex

    #2 · OpenAI

    OpenAI's coding agent, run locally through the CLI, an IDE or the desktop app, and in the cloud through ChatGPT.

    A−7/8On the 8 tests
    • Does OpenAI Codex remember across sessions?Yes
    • Can you see what OpenAI Codex remembered?Yes
    • Can you edit OpenAI Codex's memory?Partly
    All 8 answers →
  • Cline

    #3 · Cline Bot Inc.

    An open-source coding agent in your IDE, terminal and desktop, with rules and a documented Memory Bank pattern for carrying context between sessions.

    A−7/8On the 8 tests
    • Does Cline remember across sessions?Yes
    • Can you see what Cline remembered?Yes
    • Can you edit Cline's memory?Yes
    All 8 answers →
  • 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.

    B+6½/8On the 8 tests
    • Does Windsurf (now Devin Desktop) remember across sessions?Yes
    • Can you see what Windsurf (now Devin Desktop) remembered?Yes
    • Can you edit Windsurf (now Devin Desktop)'s memory?Yes
    All 8 answers →
  • Gemini CLI

    #5 · Google

    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.

    B+6½/8On the 8 tests
    • Does Gemini CLI remember across sessions?Yes
    • Can you see what Gemini CLI remembered?Yes
    • Can you edit Gemini CLI's memory?Yes
    All 8 answers →
  • Aider

    #6 · Aider-AI

    AI pair programming in your terminal, on a new or existing codebase, with any model you bring.

    B+6½/8On the 8 tests
    • Does Aider remember across sessions?Partly
    • Can you see what Aider remembered?Yes
    • Can you edit Aider's memory?Yes
    All 8 answers →
  • Cursor

    #7 · Anysphere

    An AI code editor with an Agent chat, Tab completions and rules: persistent instructions the Agent follows in every conversation.

    B6/8On the 8 tests
    • Does Cursor remember across sessions?Yes
    • Can you see what Cursor remembered?Yes
    • Can you edit Cursor's memory?Partly
    All 8 answers →
  • GitHub Copilot

    #8 · GitHub (Microsoft)

    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.

    B6/8On the 8 tests
    • Does GitHub Copilot remember across sessions?Yes
    • Can you see what GitHub Copilot remembered?Yes
    • Can you edit GitHub Copilot's memory?Partly
    All 8 answers →
The tests

8 questions, the same for every coding agent.

  1. 1

    Does a coding agent remember across sessions?

    Yes means: Persistent instructions or stored memory carry into a new session without you pasting them again.

  2. 2

    Can you see what a coding agent remembered?

    Yes means: A file or panel lists the stored memory.

  3. 3

    Can you edit a coding agent's memory?

    Yes means: The memory is editable by you.

  4. 4

    Can you delete or turn off a coding agent's memory?

    Yes means: A documented way to delete it or switch it off.

  5. 5

    Does a coding agent keep memory in a file you own?

    Yes means: The memory lives in a plain file on your machine or in your repository.

  6. 6

    Does a coding agent read a format other tools read?

    Yes means: It reads a shared convention such as AGENTS.md, so the same file works elsewhere.

  7. 7

    Does a coding agent use your code for training by default?

    Yes means: 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.

  8. 8

    What does a coding agent cost?

    Yes means: A free tier exists and the entry paid price is published.

Every grade is read from the maker's own documentation or pricing page on the day shown, with the page linked on the product's page. Nothing is graded from a review, a blog or a test we ran; where we test a product ourselves the page will say so. If a page has changed, tell us.

Read through the framework

What these grades say about continuity, and what they do not.

Every grade on this site is read from the maker's own pages. The Continuity Framework asks five things of any system that carries state: that it carry, govern, evolve, prove and forget. These tests reach carry, govern, prove and forget; they do not reach evolve, which no maker's page documents in a gradeable way. A high letter is a good coding agent, not a continuity layer.

C1 RemembersCarry
Remembering across sessions is question C1: state outlives the session that created it.
C2 See itProve
Being able to see the stored memory is the smallest form of question P1: the owner can inspect what is carried.
C3 Edit itGovern
An editable memory means the owner, not only the agent, has write authority (question G1, in part).
C4 Delete itForget
Deleting or switching off memory is rung 5 of the erasure ladder for the store the agent keeps; summaries and indexes it built are not covered unless documented. Rung 5 of the erasure ladder: delete the record.
C5 Your fileCarry
Memory in a file you own is portability by construction: question C3 answered without an export.
C6 PortableCarry
Reading a shared convention such as AGENTS.md is question C2: the same file works when the agent changes.
C7 TrainingGovern
Not training on your code by default is a scope boundary on the maker's side: what leaves the record and where it may go.
Questions people ask

The short answers.

Which coding agents keep memory in a file I own?
Claude Code, Codex, Gemini CLI, Cline and Aider keep their memory and instructions in plain files on your machine or in your repository; Cursor and Copilot keep part of it on your account, and Devin Desktop (Windsurf) keeps rules in files. Each product's page shows the exact path and the maker's own line.
Which coding agents read AGENTS.md?
Claude Code, Cursor, Codex, Copilot, Devin Desktop, Gemini CLI and Cline all state that they read AGENTS.md; Aider's docs do not mention it. One file can therefore serve most of them.
Does my coding agent train on my code?
It depends on the maker and the plan. Copilot's individual plans have a training setting you can disable; Aider collects nothing; Claude Code, Cursor, Codex and Gemini CLI say it depends on your plan or setting. Each product page quotes the maker's own line and links to it.

Placement here is never for sale and no link on this page pays us. How links work on this site. The consumer assistants are graded the same way on the Scorecard.