What is AideMemo?

AideMemo is a local memory layer for coding agents and developer tools. It stores facts, decisions, lessons, and errors in a single local database, then exposes that memory through a CLI, MCP tools, the agent SDK, and native bindings.

The default memory path covers capture, typed writes, BM25-first search, and MCP or agent SDK reads. It runs locally without an external LLM call. Remote extraction, embedding, and reranking remain opt-in.

Use it when your agent needs project memory that survives across sessions, editors, and model providers.

Shared memory and handoff solve different parts of that continuity. Shared memory keeps durable context available to connected agents. Handoff deliberately moves one tracked task to a named worker with a focus, a done_when condition, and a result linked back to the same session.

For the system map, read Architecture. To run one durable store across multiple agents, read Shared Memory Layer. To move task ownership, read Hand off a tracked task. For the validated scorecard, read Evidence. For the per-turn tool choice guide, read Agent Workflows.

What it gives you

Need	AideMemo feature
Remember project decisions	aidememo fact add --type decision
Search prior context	aidememo search and aidememo query
Start work from a sparse ticket	aidememo workflow start
Share memory across agents	Stable source_id / actor_id identities and aidememo mcp-serve
Move a tracked task to another coding-agent account	aidememo handoff send ALIAS
Give agents tool access	aidememo mcp or HTTP MCP
Use memory from code	aidememo-agent-sdk

Basic model

AideMemo stores three main things:

• Entities: topics such as Redis, Billing, Codex, or a session id.
• Facts: typed memory attached to entities.
• Relations: graph edges between entities.

Facts can be typed so the right memories rank higher:

Fact type	Use for
decision	A choice that should guide future work
lesson	Something learned from a past attempt
error	A failure mode to avoid
preference	A user or team preference
pattern	A recurring architecture or workflow pattern
note	General context
question	Incoming issue, ticket, or open investigation

Typical workflow

1. Install the aidememo CLI.
2. Add facts while you work.
3. Search/query memory before making a plan.
4. Register AideMemo as an MCP server for your agent.
5. Use workflow start for issue, PR, or ticket automation.
6. When ownership changes, use handoff send so the next worker continues the same tracked session and returns evidence.

aidememo fact add \
  "Decision: Redis timeout fixes must go through the Worker job wrapper." \
  --type decision \
  --entities Redis,Worker

aidememo query "Fix Redis timeout in worker"

What AideMemo is not

AideMemo is not a hosted memory service, a full agent runtime, or a replacement for your issue tracker. It is a local memory system that your existing tools can call.

If you want a cloud-managed agent platform, use a hosted memory/runtime product. If you want explicit local memory with a CLI, MCP tools, and SDK access, use AideMemo.