Contracts and routes
Use this section when you already understand the product shape and need route-level or SDK-level detail.
This section is the bridge between product docs and raw repository contracts. Use it when you need to know which public surface exists, which route backs it, and where the literal source of truth lives.
Memory
/v1/memory/*Write, recall, planning, task start, lifecycle reuse, sessions, tools, and review helpers.
Cross-cutting surfacePolicy Memory
feedback -> experience -> review -> governancePersisted policy contracts, evolution inspect, and the retire/reactivate governance path.
Route familyHandoff
/v1/handoff/*Pause and resume through explicit handoff packets and anchor-based recovery.
Route familyReplay
/v1/memory/replay/*Run lifecycle, playbooks, promotion, dispatch, repair, and local reuse.
Route familyReview
/v1/memory/*review*Continuity review packs, evolution review packs, and replay repair review.
Route familyAutomation
/v1/automations/*Validate, create, run, inspect, resume, and cancel Lite automation graphs.
Route familySandbox
/v1/memory/sandbox/*Create sessions, execute bounded commands, and inspect logs and artifacts.
Think in route families first, not single endpoints. Most integration mistakes happen when teams read one method in isolation and miss the surrounding continuity loop.
Best entry points
Start with the page that matches the runtime surface you are integrating:
- Memory
- Policy Memory and Evolution
- Handoff
- Replay and Playbooks
- Review Runtime
- SDK Quickstart
- Lite Config and Operations
- Automation
- Sandbox
Memory
Write, recall, planning, task start, sessions, tools, review packs, and delegation-learning helpers.
Cross-cutting surfacePolicy Memory and Evolution
Read how feedback can materialize policy memory, how review exposes evolution inspect, and how governance retires or reactivates learned policy.
Reference familyHandoff
Structured pause and resume, task anchors, and the handoff path into host-task sessions.
Reference familyReplay and Playbooks
Replay lifecycle, playbook compilation, promotion, repair, and Lite execution reuse.
Reference familyReview Runtime
Continuity review packs, evolution review packs, and replay repair review in the Lite governed subset.
How to use this section
Use this section in two passes:
- stay in the docs site for the public surface and family-level understanding
- only open raw contracts when you need exact field names or source-level debugging
For most integrations, you should not need to start from raw TypeScript files.
Public route families
| Family | Main job | Start here |
|---|---|---|
| Memory | Write, recall, planning, task start, sessions, lifecycle | Memory |
| Action retrieval | Retrieve the next move explicitly, with evidence, uncertainty, and gate surfaces | Action Retrieval |
| Policy memory | Persist, inspect, and govern self-evolving policy state | Policy Memory and Evolution |
| Handoff | Pause, store, recover, and resume task state | Handoff |
| Replay | Record runs, compile playbooks, promote and reuse | Replay and Playbooks |
| Review | Build continuity/evolution review material and review replay repairs | Review Runtime |
| Automation | Validate, create, run, pause, and resume local automation graphs | Automation |
| Sandbox | Execute bounded local runtime actions and inspect logs/artifacts | Sandbox |
| Runtime operations | Boot, health, config, Lite boundary | Lite Runtime |
| SDK integration | Call the runtime from TypeScript | SDK Quickstart |
Common lookup tasks
| If you are trying to answer... | Read this page first |
|---|---|
| "How do I ask for a better first action?" | Memory |
| "How do I ask what the agent should do next, and why?" | Action Retrieval |
| "How do I know when the runtime wants inspect / widen recall / rehydrate first?" | Uncertainty and Gates |
| "How do I turn gate output into a host action?" | Operator Projection and Action Hints |
| "How do I inspect or govern self-evolving policy state?" | Policy Memory and Evolution |
| "How do I pause work and resume it later?" | Handoff |
| "How do I turn a successful run into something reusable?" | Replay and Playbooks |
| "How do I validate or run an automation graph?" | Automation |
| "How do I execute bounded local commands and inspect outputs?" | Sandbox |
| "How do I get review-ready runtime state?" | Review Runtime |
| "Which route families are really in Lite?" | Lite Runtime |
| "How do I integrate this from code?" | SDK Quickstart |
What the public docs already cover
The docs site already explains:
- the runtime model
- the main continuity surfaces
- the explicit action-retrieval and gate surfaces
- the public SDK integration path
- the Lite runtime boundary
- the most important memory, handoff, and replay families
That means the raw repository references should be the second stop, not the first one.
Use the docs site for shape and meaning. Use raw contracts only when you need exact field names, support boundaries, or source-level debugging.
Raw sources when you need exactness
Sometimes you do need the literal source of truth. Use the raw sources below only when the docs page above is not enough.
SDK contracts
The main typed SDK request and response shapes live here:
Use that file when you need:
- memory request shapes
- replay payload shapes
- handoff envelopes
- automation and sandbox contracts
Runtime capability matrix
The best raw route-level reference today is:
Use it to answer:
- which routes are supported
- which are Lite subsets
- which are conditional
- which are intentionally unavailable in Lite
Runtime boundary
For public product boundary and open-core distribution stance:
Package-level references
How to use this page
Use this page for the public shape and integration path first.
When you need exact field-level detail, continue into:
- the repository contract files
- the capability matrix
- the SDK contract sources