Your agent remembers.
And your team's agents too.
Auto-captured, auto-recalled, path-scoped memory for AI coding agents and teams.
npm install -g aide-memory && aide-memory initThe gap aide-memory closes
Static rules files (CLAUDE.md, .cursorrules) and skills cover always-on guidance and procedures. They're a starting point, but they don't do scoped, layered, auto-captured project memory. aide-memory adds that layer.
Rules files don't scale
One flat file, no path scoping, no categorized layers, manual to update. The whole file gets injected on every turn, crowding context with things that don't apply to the area the agent is working in.
Cross-session and cross-teammate gaps
The fix you taught your agent yesterday isn't in today's session by default. When a teammate opens the same area, their agent starts fresh too, with little of what you learned carrying over.
Tool-specific memory
You can copy rules files between tools manually, but what you teach your agent in one tool doesn't carry to the other on its own. Switch tools or hand off to a teammate on a different tool and your project context stays behind.
Manual capture, unscoped recall
Capturing context means manually editing a rules file. Recall is unscoped: the whole file gets injected every turn whether it's relevant or not. aide-memory automates capture via hooks and scopes recall to the area the agent is working in.
How it's built different
Prompted at the right moments
When the agent opens a file with relevant memories, it gets prompted to recall them. When you correct the agent, the hook detects it and prompts it to store the lesson. Decisions and findings surface via periodic reflection.
Scoped, layered recall
Memories attach to the code area they apply to (src/auth/**, packages/api/**) and to one of four categorized layers. The agent gets back what's relevant to that scope, not the whole rules file.
Git-synced for teams
Memories are JSON files in your repo. Commit, push, pull, your teammates' agents pick up the same context. Private preferences stay gitignored.
Project memory, not tool memory
Memories belong to your project, not to whichever tool you're using. Switch tools or hand off to a teammate and the memory store comes with the codebase.
Nudge, don't dump
A small nudge tells the agent memories exist for the path it just opened. The agent decides whether to pull them. A small rules file is injected per turn so the agent knows how to use the tools, but memory content comes in on demand, scoped to the area.
Tunable to your flow
Defaults capture the common case. Prompt cadence, recall scope, injection caps, and search mode are all tunable.
Local-first storage
Memories live as JSON files in your repo plus a local SQLite cache. Code and memory content never leave your machine. Anonymized usage counts ship by default; disable with AIDE_TELEMETRY=off.
Uses your existing agent
aide-memory does no LLM calls of its own. The model in your editor does the reasoning. The overhead is ~2,900 tokens for aide-memory's seven MCP tool definitions (loaded once per session) plus the memories the agent pulls back per area.
Free
Free to use. More features or tiers may be added.
How it works
Init
npm install -g aide-memory && aide-memory initCreates .aide/, installs six hooks, configures the MCP server, writes Claude Code + Cursor rules files. Seconds.
Capture as you work
Corrections are detected automatically and the agent gets prompted to store them. A periodic reflection at the end of turns picks up decisions and findings. Session tracking resets before context compaction so recall stays clean.
Recall in your next session, or your teammate's
8 memories exist for src/checkout/**. Call aide_recall if relevant.The hook prompts aide_recall and the agent pulls back what applies to that area. Commit .aide/memories/ and your teammates' agents pick up the same context on their next read.
Stay in the loop
Occasional updates on aide-memory: new editor adapters and releases.
Get started in seconds.
Free. Local-first. No account required.
npm install -g aide-memory && aide-memory init