Skip to main content
AideMemo logoCross-agent project continuity

AideMemo

Switch coding agents. Keep the work moving.

Carry decisions, failed attempts, and lessons across Hermes, Codex, Claude Code, and other coding agents. No chat sync. No restart from zero. The default local memory loop does not require an external LLM call.

Give my agents memoryTry a handoff
Install AideMemocurl -fsSL https://raw.githubusercontent.com/taeyun16/aidememo/main/scripts/install.sh | bash

Measured, with boundaries

Evidence before adjectives.​

Public claims stay attached to a dataset, retrieval stack, and execution envelope.

Read the scorecard
0.992R@10

LongMemEval-S, 500 questions, opt-in BGE plus two-stage rerank

5.7×faster

Daemon BM25 versus fresh CLI, with the same BrainBench score

18.4ms p50

Shared HTTP MCP, two clients and twenty persisted writes

One project. Any agent.

The conversation ends. The work does not.​

Shared memory keeps project context available. Handoff deliberately changes ownership with a focus, a definition of done, and returned evidence linked to the same session.

Hand off a tracked task
SENDER · HERMESNames the next objectiveReview the refresh-token fix and run focused tests.
HANDOFF · SAME SESSIONCarries the durable evidenceFocus, done_when, decisions, failures, and lessons.
RECEIVER · CODEXContinues and returnsThe result fact lands on the same tracked workflow.

The local memory loop

Memory that stays close to the work.​

Capture explicitly, retrieve deliberately, and expose only the context the next task needs.

01Capture

Keep the facts that should outlive this session.

Decisions, lessons, errors, preferences, and relations stay explicit instead of disappearing into a transcript.

aidememo fact add "Use Redis Cluster" \ --type decision --entities Redis,Cache
02Retrieve

Recall the right context before the next plan.

BM25-first search stays fast and deterministic, while semantic retrieval steps in when lexical evidence is weak.

aidememo query "Redis cache" --depth 2
03Apply

Bring memory into the tool already doing the work.

Use the same local store through MCP, the agent SDK, the CLI, or native bindings without adopting a hosted runtime.

aidememo init --agent codex ./wiki

One core, several entry points

Meet the agent where it runs.​

MCPModel-visible tools over stdio or HTTPAgent SDKCode-first memory composition from PythonCLICompact commands for humans and automationNativePython, Node, Elixir, and C bindings

Documentation

Go from first fact to full integration.​

English and Korean guides share the same tested routes, examples, and release gates.

01Start hereUnderstand the product boundary and the basic memory model.02QuickstartCreate a store, add typed facts, and recover them from a ticket.03Hand off a tracked taskSend one session to a named coding-agent account and link the returned evidence.04ArchitectureSee how the interfaces, Rust core, stores, and indexes fit together.05Coding agent setupInstall AideMemo for Claude Code, Codex, Hermes, pi, and other agents.06MCP setupRegister AideMemo with local agents and use the core tools.07Shared memoryGive multiple agents one durable store with explicit source and writer identities.08Agent workflowsChoose context, query, workflow, and aggregate calls by task shape.09Multiple Codex accountsShare one project memory while login state and writer provenance stay separate.10EvidenceRead the measured outcomes, caveats, and claim boundaries.

AIDEMEMO / 0.1

Your project memory belongs to the project. Not the agent.​

Give my agents memoryView on GitHub