The Continuity Layer
An independent public map of AI continuity, memory, and stateful agent systems — maintained from ongoing work in this field.

AI needs more than intelligence. It needs continuity.

Eight co-equal axes. No center, no hierarchy.

14 Field categories07 Signals tracked
MemoryContextStateProvenanceAgentsWorkflowsEvaluationInfrastructure

The continuity field, mapped along eight co-equal axes. Select any to see what it covers.

01The problem

Why continuity matters

AI interactions are powerful but volatile — that’s why AI keeps forgetting. Every new session, tool, and model can drop the thread; context disappears, drifts, or becomes unsafe when it is not governed.

Problem

AI interactions are powerful but volatile; context disappears, drifts, or becomes unsafe when it is not governed.

Mission

The Continuity Layer is the durable responsibility for deciding what carries forward across models, tools, agents, and sessions.

Thesis

Memory stores. Retrieval finds. Agents act. Runtime executes. Continuity governs what carries forward.

  1. 01

    The problem isn’t memory. It’s everything around it.

    Storing facts is largely solved. What still evaporates between sessions is state: decisions, the reasons behind them, working assumptions, unresolved threads, and where each piece of truth came from.

  2. 02

    Continuity is the layer that carries understanding forward.

    It’s what lets intelligence keep what it has learned, decided, inferred, promised, questioned, and left unfinished — across chats, tools, agents, and time.

  3. 03

    Agents make this urgent.

    Every new orchestrator spawns more carriers of partial memory and stale assumptions. More agents means more coordination problems — unless they converge on one shared account that is authoritative for that project's governed state, so people and agents build on the same record instead of coordinating with each other.

  4. 04

    The problem predates AI.

    People leave, decisions get buried, documents go stale. Decades of partial answers — wikis, decision records, audit trails, commit lineage — are evidence that the problem is real. AI didn’t create it; AI multiplied it.

  5. 05

    A field is forming, and we haven’t found an independent, vendor-neutral map of it being kept up to date yet.

    Memory systems, context engineering, stateful agents, provenance, evaluation — the pieces are emerging faster than the vocabulary. One underexplored question is whether long-running AI systems need project-level continuity in addition to agent-level infrastructure. This site exists to track that question in public.

02The definition

A continuity layer is the part of an AI system that carries forward governed state across sessions, tools, and models: what is currently true, what changed, why it changed, and which evidence supports it. It sits above memory and retrieval, and below agent execution, turning stored history into usable, auditable continuity.

03New here

Choose the path that matches what you need

If you are arriving fresh, start with the level of explanation you need before moving into the map and evidence.

04Stack position

Layer, Engine, Runtime

The Continuity Layer is the durable frame that decides what carries forward. The Continuity Engine implements that frame, and the Continuity Runtime is where users, models, tools, and agents execute the work.

System model

A governed path from stored history to current working truth.

  1. 01LayerDecidesSets the durable frame for what should carry forward.
  2. 02EngineGovernsSupersedes stale facts, keeps lineage, and reconstructs current state.
  3. 03RuntimeSurvivesCarries that governed state through real model, tool, and agent work.
Layer01

Durable frame

The field frame and architectural responsibility for deciding what should persist across models, tools, agents, and sessions.

what carries forward
Engine02

Implementation

The implementation that governs writes, supersedes stale facts, keeps lineage, and reconstructs the current account of the work.

how it is governed
Runtime03

Execution space

The live execution space where the user, model, tools, and agents interact while the Engine governs the work.

where it must hold
Runtime flow
  1. User
  2. Continuity Runtime
  3. Model / Tool / Agent
  4. Continuity Runtime
  5. User
Above

Agent execution — the tools and agents doing the work

The continuity layer

Governs state: what is true now, what changed, why, and the evidence behind it

Below

Memory & retrieval — where facts are stored and fetched

05Core components

What a continuity layer is made of

The eight axes are not a second framework. They are the field dimensions the Layer reads and the Engine governs into a single working account across time.

Memory

What is stored.

Memory systems, Personal AI memory

Context

What is available right now, in this call.

Context engineering

State

What persists — the current shape of a project or task.

Stateful agents, Long-running workflows

Provenance

Where each piece of truth came from.

Provenance & audit, Organizational memory

Agents

Who acts, and under what authority.

Agent infrastructure

Workflows

How continuity survives a handoff — session to session, tool to tool, agent to agent.

Long-running workflows, MCP & tool continuity

Evaluation

How any of this is measured, and against what standard.

Eval benchmarks, Papers

Infrastructure

Where the layer actually lives — protocol, runtime, product, or process.

MCP & tool continuity, Commercial

How these differ from continuity — the Primer →
06Field evidence

The field, mapped

The Continuity Layer maintains a public map of the emerging systems that let AI preserve context, state, memory, provenance, and working understanding across time.

A preview, not the full record — tracked signals, map infrastructure, and the Lab path where Engine behavior has its first recorded evidence.

Eight co-equal axes of the field — no center, no hierarchyMemoryContextAgentsEvaluationProvenanceStateWorkflowsInfrastructure

The continuity field, mapped along eight co-equal axes. Select any to see what it covers.

14Field categories
7Signals tracked

These surfaces show the field forming: categories, signals, and evidence infrastructure. The Lab holds the first recorded evidence of Engine behavior — Demo 001, a PASS bounded to read-only orientation and conflict rejection.

07Open Questions

Open questions the map is holding

Questions this project tracks rather than claims to have answered. Drafted, and still moving.

  1. 01

    What should persist across a project’s life — and what is noise a system should be free to forget?

  2. 02

    Who owns a project’s memory when a person, an app, a model, and an agent all touch it?

  3. 03

    How should an agent inherit context it didn’t create without inheriting its mistakes?

  4. 04

    How should provenance be represented, so any decision can show where it came from?

  5. 05

    What should stay portable across models, and what is tied to the system that made it?

  6. 06

    When two agents hold conflicting versions of the truth, on what basis does one win?

09Contribute

Found a signal?

Know a lab, tool, paper, or essay that belongs on the map and isn’t here yet? Send it. Not everything gets added — but every signal gets read.

Stay on the map

The field moves weekly. The map keeps up.

Get new Field Watch entries, Lab demos, and guide updates. No spam, no course pitch, no fake urgency. Unsubscribe anytime.

Disclosure

Maintained by an independent builder testing continuity patterns in live systems.

This map is maintained by an independent builder who tests continuity patterns in ongoing work in this field. The maintainer also does commercial work in this field. The two are kept separate: the map sells nothing, takes no sponsorships, and favors no vendor — including the maintainer’s own.

The Continuity Layer is maintained by an independent builder testing continuity patterns in live systems. Entries reflect what can be verified from papers, products, and production use — claims are labeled as verified or speculative, and the map is corrected as the field moves.