update
May 1, 2026
By Teun
aide-memory adds scoped memory for coding agents and teams
aide-memory is a new tool for AI coding agents that stores corrections, technical context, and team guidelines in scoped JSON memories. According to the project, it works with Claude Code and Cursor, syncs through git, and keeps memory content on the user’s machine.
aide-memory is a new persistent memory tool for AI coding agents and teams. The project, announced on April 29, 2026, is designed to capture corrections, technical context, and area-specific decisions while developers work, then recall only the parts that apply to the file or code area being edited.
The creator says the tool exists because rules files such as CLAUDE.md and .cursorrules are useful, but limited. According to the post, those files do not automatically retain corrections made during a session, they do not scope knowledge to a specific code area, and they inject the same context on every turn whether or not it is relevant.
⚡ New to this?
This is news because it tries to solve a common problem with AI coding assistants: they forget what you told them in earlier sessions. Rules files like CLAUDE.md are static instruction files, while aide-memory adds stored context that can be recalled later and scoped to specific parts of a codebase.
The main idea is that the agent can remember team habits, module decisions, and corrections without users retyping them. That matters for people working in larger codebases, where a setting panel, API package, or dashboard can each have different conventions.
🦞 OpenClaw angle
If you run self-hosted coding agents, test whether your current prompt rules are doing too much work that should be stored as scoped memory instead. Use separate memory layers for team-wide rules, module-specific decisions, and personal preferences so your agent does not get a giant global prompt on every turn.
If you work in a team, keep shared memory under version control and keep personal preferences gitignored, as the project does. Also check whether your automation can trigger recall before file edits, because the point here is to surface context only when the agent is actually working in that path.
With aide-memory, capture happens automatically through hooks installed by running aide-memory init. The tool uses six hooks across the session lifecycle, including a UserPromptSubmit hook that watches for corrections, a Stop hook that periodically asks whether anything should be remembered, a SessionStart hook that injects top preferences and team guidelines, and a PreToolUse hook that checks for memories relevant to the file being opened or edited.
The project stores memory in JSON files under .aide/memories/ and uses a local SQLite cache. Memories are organized into four layers: preferences, technical, area_context, and guidelines. According to the documentation in the post, recall prioritizes area_context first, then technical, then preferences, then guidelines.
That scoping is central to the tool’s design. Instead of sending a global block of text to the agent on every prompt, aide-memory matches the current file path against stored scopes such as src/components/dashboard/ or packages/api/ and surfaces only the relevant memories.
The author says the tool also supports team workflows. Memories can be committed to git, pushed, and pulled so that a teammate’s agent can pick up the same context after a post-checkout hook rebuilds the local cache. Personal preferences are gitignored, while team conventions can travel with the repository.
The tool is meant to work across editors as well. The post says Claude Code and Cursor both read the same .aide/memories/ directory, so context captured in one tool is available in the other. It also notes that Codex, Copilot, and Windsurf currently get a rules template at launch, with deeper integration possible later.
Aide-memory is configurable. The project says users can tune how often the agent reflects, how specific a scope must be before memory appears per file, how much context gets injected at session start, and which hooks are active. aide-memory init seeds .aide/config.json with public settings so they can be edited in one place.
According to the post, privacy is handled locally: memory content stays on the user’s machine, with telemetry limited to event types and machine-anonymous counts unless turned off through AIDE_TELEMETRY=off or aide-memory config telemetry.enabled false. The project is free to use and can be installed with npm install -g aide-memory followed by aide-memory init.
