Recall

Stop explaining your codebase to AI every new session.

Recall scans a repository and generates persistent, evidence-backed context that Claude Code, Codex, Gemini CLI, Cursor, and other coding agents can reuse.

Early public release. Commands, output format, and internals may still change between versions. See Current scope for what's supported today.

Quick start

npx recall-context init

npx recall-context context --stdout

Or install it once and use the shorter recall command:

npm install -g recall-context
recall init

The npm package is recall-context; it installs the recall and recall-context commands, which invoke the same CLI.

Requirements

• Node.js 22 or newer.

What it generates

.recall/
├── manifest.json       # pointer to the current snapshot + memory files
├── architecture.md      # detected apps/packages/layers, entry points, dependency direction
├── conventions.md       # naming, aliases, linting/formatting, scripts
├── decisions.md         # human-editable decision log
├── features.md          # features backed by structural evidence (controllers, routes, jobs...)
├── glossary.md          # recurring domain terms extracted from identifiers
├── risks.md              # evidence-based risk findings (missing tests, circular deps, etc.)
├── technical-debt.md    # measurable debt findings with evidence and severity
└── snapshots/
    └── latest.json       # the deterministic scan result everything above is derived from

Every generated file wraps Recall's content in markers; anything you write outside them is yours and survives every recall update.

Before / after

Before, every new AI session:

New session → explain architecture → explain conventions → explain important files → repeat next session

After:

recall context --stdout

Paste or pipe the output into your coding agent.

Recall vs CLAUDE.md / AGENTS.md

| | CLAUDE.md / AGENTS.md | Recall | | ------------------- | ----------------------------- | -------------------------------------------------- | | Authored by | A human, by hand | Derived from repository evidence | | Stays current? | Can go stale silently | Detects and reports staleness (recall status) | | Shape | Typically one static document | Snapshot + multiple focused files | | Scans the repo? | No | Yes — deterministic, local scan | | Evidence paths | Not by default | Every claim traces back to source | | Works across agents | Depends on the file | Same output for any agent that reads Markdown/JSON |

Recall doesn't replace CLAUDE.md or AGENTS.md. Use CLAUDE.md/AGENTS.md for human-authored instructions — things only a person should decide. Use Recall for generated repository context and freshness checks — things that can be derived and kept honest automatically. Most projects will want both.

Commands

recall init                                  # analyze the repository and create .recall/
recall scan                                   # re-run deterministic analysis, save a snapshot
recall update                                 # refresh memory from the current repository state
recall status                                 # check whether memory is initialized and up to date
recall explain <path>                         # explain a file or directory's role
recall context                                # generate compact, agent-ready context
recall context --task "Add password reset"    # focus that context on a specific task
recall doctor                                 # validate the environment and memory state

See docs/cli-reference.md for every flag and exit code.

Safety

• Local-first: everything runs on your machine, in your repository.
• No account, sign-up, or API key required.
• The CLI sends no telemetry. The documentation website is a separate application with its own operational analytics.
• No AI provider is required or called by default (see docs/provider-interface.md).
• Does not execute, import, or evaluate your project's source code or scripts.
• Managed writes are restricted to .recall/, .gitignore, and any output path you explicitly pass.
• Refuses to write through a symlinked .recall directory or a symlink escape anywhere in the write path.
• Generated Markdown preserves human-authored content outside its <!-- recall:generated:* --> markers.

Current scope

Supported now: JavaScript, TypeScript, Node.js, NestJS, Next.js, pnpm/npm/Yarn, monorepos, both Git and non-Git repositories.

Not yet built: an MCP server, additional language ecosystems, deeper semantic (AST-based) analysis, an optional AI-assisted "why" layer, editor integrations. See docs/roadmap.md for the full list and what's deliberately deferred.

Demo

A 20–30 second terminal recording is planned but not yet produced — see docs/demo-script.md for the exact script and docs/assets/ for where the recording will live once made.

Learn more

• CLI reference — every command, flag, and exit code
• Memory format — the on-disk .recall/ format
• Architecture — how the monorepo is structured
• Roadmap — what's in scope, deferred, and planned
• Changelog
• Contributing
• Security policy
• npm package
• Releases
• Issues

A note on the name

Recall's longer-term vision includes tracking why decisions were made, not just what the code currently looks like. That's a future direction, not a current capability — today decisions.md records only what's structurally evidenced plus whatever a human writes into it by hand.

License

MIT