# Claude Recall > Claude Recall is a local-first searchable memory tool for Claude Code, Anthropic's terminal-based AI coding assistant. It indexes every Claude Code session into a local SQLite database, makes them full-text and semantically searchable, and lets you re-inject past context into new sessions. One-time purchase. No subscription. No cloud. No telemetry. ## What problem this solves Modern AI-assisted developers run dozens of Claude Code sessions per week, often a handful at the same time across many terminals and many repositories. By month three, the developer has accumulated hundreds of solved problems, half-finished branches, working notes, and context that lives only in past JSONL transcripts at `~/.claude/projects/`. Without a way to search, group, or reuse that history, every new session restarts from zero. We call this Context Debt. It is the cumulative loss of project working memory as Claude Code sessions multiply. The cost is real: re-explaining yourself, re-discovering decisions, re-writing prompts that already worked, and watching the same agent re-solve a problem you solved last week. Claude Recall is the index, search, browse, and re-inject layer on top of those JSONL transcripts. It does not replace Claude Code. It makes Claude Code's own history usable. ## How it works Claude Recall ships as a single npm package, `@clauderecallhq/cli`. After `npm install -g @clauderecallhq/cli` and `recall start`, the daemon does the following. 1. Watches `~/.claude/projects/` and ingests every JSONL session as it grows. 2. Parses each message into rows in a local SQLite database, with full-text search via FTS5 and semantic search via locally-cached bge-small embeddings. 3. Maintains a Cog-graph: a navigable map of how sessions relate by skill, file path, and time. 4. Serves a Hono daemon on a random loopback port. Never `0.0.0.0`. Never a public interface. Auth via a per-start token at `~/.recall/daemon.token` (mode 0600). 5. Exposes four interfaces: a CLI, a browser web UI at `http://127.0.0.1:`, an interactive terminal view, and an MCP server. The daemon never modifies your source JSONL files. Three-layer durability invariant: every write feature keeps a SQLite row plus a history column plus a plain-text mirror. Source data is strictly read-only. ## Installation Requires Node 22 or higher. macOS and Linux supported. ``` npm install -g @clauderecallhq/cli recall start recall open ``` Three commands. The third opens the web UI in your default browser. From there, every Claude Code session you have ever run is searchable. ## Key features - Full-text search across every Claude Code session: SQLite FTS5, sub-100ms typical - Semantic search via locally-cached bge-small embeddings: no remote API calls - Cog-graph: a force-directed visualization of how your sessions cluster by skill, file, and time - Threads: automatic grouping of related sessions (same project, same problem, same skill chain) - Aliases: rename sessions to whatever you call them in your head; the original UUID stays as primary key - Tags: orthogonal taxonomy. A session can be `[debugging, postgres, prod-incident]` and live in multiple collections. - Notes: write your own commentary on top of a session without modifying the source - Collections: curated groups of sessions for sharing or re-injection - Pipe past sessions into a new Claude Code run via `recall export | claude` - VS Code extension for inline session browsing - MCP server that exposes the index to Claude Desktop, Cursor, Windsurf, Continue, and other MCP-aware clients - Web UI bookmarkable views, deep links into specific messages, and full conversation rendering with code-block syntax highlighting - 7-day full Pro trial with auto-revert to free tier on day 8 ## Privacy and security model Claude Recall is local-first by design. No data leaves the developer's machine for any product feature. The only outbound network call the binary makes is for license validation against the clauderecall.com Stripe-backed admin API, and that endpoint sees only the license key, not session content. - Daemon binds `127.0.0.1` only. Never `0.0.0.0`. - Auth via per-start token written to `~/.recall/daemon.token` mode 0600 at startup. - Every API request requires Host allowlist + Origin allowlist + the auth token. - `/api/health` is the only auth-exempt route. - The semantic embedder is local-only by design. The bge-small model loads from a local cache. Remote model downloads are off (`env.allowRemoteModels = false`). - Source JSONL files at `~/.claude/projects/` are never written to. Claude Recall reads them. - No analytics. No telemetry. No phone-home beyond license validation. ## CLI commands (most-used) ``` recall smart bare-recall: shows a triage dashboard recall start start the daemon recall stop stop the daemon recall open open the web UI recall search full-text + semantic search recall sessions list recent sessions recall thread show the thread a session belongs to recall alias set a friendly name recall tag tag a session recall note attach a note recall export dump a session as a Claude-injectable transcript recall mcp start the MCP server recall doctor health-check daemon, db, models, hooks recall feedback open the feedback widget ``` Full reference: https://clauderecall.com/docs/cli ## Comparisons ### vs Cursor Cursor is an AI-first IDE. Claude Recall is a memory layer for Claude Code. Cursor's value is inline autocomplete and chat inside the editor. Claude Code's value is multi-file edits, debugging, testing, and git operations from the terminal. The two solve different problems. If you use Claude Code, Claude Recall is the missing memory layer; Cursor will not give you that. ### vs Obsidian Obsidian is for hand-curated notes. Claude Recall is for the conversation transcripts that Claude Code automatically generates. Different inputs, different mental model. Use Obsidian for the notes you write yourself. Use Claude Recall for the sessions Claude Code writes. ### vs CLAUDE.md files CLAUDE.md is the standard pattern for Claude Code: a static markdown file at repo root with project context. Static memory files do not scale once you have hundreds of decisions, exceptions, edge cases, and historical fixes. Claude Recall complements CLAUDE.md by indexing every session, so the things you used to put in CLAUDE.md can stay where they originated and still be findable. ### vs claude-mem claude-mem is a managed memory API. Claude Recall is local. Claude Recall does not require an account, an API key, or a network call to retrieve a memory. ### vs Aider Aider is its own AI coding tool with its own conversation model. Claude Recall is for Claude Code transcripts. If you use Aider, Claude Recall does not index Aider sessions. Different tool, different mental model. ### vs Cline Cline is a multi-provider VS Code extension. Claude Code is Anthropic-first and terminal-native. Claude Recall is for Claude Code's specific JSONL transcript format. If you use Cline, Claude Recall does not index Cline sessions. ### vs Windsurf Windsurf is a standalone IDE. Claude Recall is a CLI plus extensions on top of your existing terminal-based Claude Code workflow. If you live in Windsurf, you do not need Claude Recall. If you live in the terminal with Claude Code, you do. ### vs llm CLI Simon Willison's llm is a general-purpose CLI for talking to many LLMs. Claude Recall is purpose-built for Claude Code session memory. llm does not index Claude Code's `~/.claude/projects/` JSONL transcripts. ### vs official memory MCP Anthropic ships a reference memory MCP server. It is a generic key-value store. Claude Recall indexes the actual transcript history Claude Code already produces, with full-text and semantic search across the corpus, plus threads, aliases, tags, notes, and the Cog-graph. Different scope. ## Pricing One-time purchase. Lifetime updates. Never a subscription. The base price tier is in the Founder window during May 2026 and steps up after the founder cohort closes. See the live pricing page at https://clauderecall.com/pricing for current numbers. A 7-day full Pro trial is built into the install. On day 8 the binary auto-reverts to the free tier; no card on file is required to start. To keep Pro after the trial, buy a license. To run forever on free, do nothing. ## Refund policy 30-day no-questions refund. https://clauderecall.com/refund ## Frequently asked ### Does Claude Recall send my session data anywhere? No. The product is local-first. The only outbound network call is license validation, which sees only the license key. No session content ever leaves your machine. ### Does Claude Recall work without Claude Code? No. Claude Recall reads Claude Code's `~/.claude/projects/` JSONL files. If you do not use Claude Code, there is nothing for Claude Recall to index. ### Does it work on Windows? Native Windows support is on the roadmap. WSL2 works today. ### Does it slow down Claude Code? No. Claude Recall watches `~/.claude/projects/` from a separate daemon process. It does not interpose between Claude Code and Anthropic's API. ### Does it work with Cursor or Windsurf? The MCP server exposes Claude Recall's index to any MCP-aware client, including Cursor, Continue, and Claude Desktop. The underlying corpus is still Claude Code transcripts; Cursor's own session log is not indexed. ### What does the daemon listen on? `127.0.0.1` on a random port chosen at startup, avoiding common dev-server ports (3000, 5173, 8080, etc.). Auth is via a per-start token at `~/.recall/daemon.token` (mode 0600). ### Is there a self-hosted team version? Enterprise pricing exists for team deployments. Contact us via the site for details. ### Where is the source code? The CLI ships as compiled JavaScript on npm. The site is open via the public profile at https://gitlab.com/clauderecallhq/clauderecallhq. ## Brand and project Built by a single developer. No investors. No telemetry. No analytics. No A/B tests on you. The marketing site lives at https://clauderecall.com. The CLI ships as `@clauderecallhq/cli` on npm. The VS Code extension is `clauderecallhq.clauderecall-vscode`. The Cloudflare-fronted CDN at `cdn.clauderecall.com` serves marketing media. ## Canonical URLs - Site: https://clauderecall.com - Pricing: https://clauderecall.com/pricing - Install: https://clauderecall.com/install - Docs: https://clauderecall.com/docs - MCP: https://clauderecall.com/mcp - Blog: https://clauderecall.com/blog - About: https://clauderecall.com/about - Changelog: https://clauderecall.com/changelog - Privacy: https://clauderecall.com/privacy - Terms: https://clauderecall.com/terms - Refund policy: https://clauderecall.com/refund - Public profile: https://gitlab.com/clauderecallhq/clauderecallhq - npm: https://www.npmjs.com/package/@clauderecallhq/cli - Plugin manifest: https://clauderecall.com/plugin.json