# Claude Recall

> Claude Recall is a local-first searchable memory tool for terminal coding agents. It indexes every Claude Code
> session and every OpenAI Codex CLI 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 sync. No silent telemetry.

Claude Recall solves the problem of "Context Debt", the cumulative loss of working context as agent sessions multiply across days, projects, and feature branches. Buyers are developers who run dozens of parallel Claude Code and Codex CLI sessions across many terminals and many repositories.

The product ships across four surfaces from a single npm install: a CLI (`recall`), a local web UI served by a Hono daemon on a random loopback port, a terminal interactive view, and an MCP server that exposes Claude Recall's index to Claude Desktop, Cursor, Windsurf, Continue, and any other MCP-aware client. There is also a VS Code extension for inline session browsing.

Both agents are indexed by default, with no configuration. Codex CLI sessions land beside Claude Code sessions in the same database and the same search, wear a `Codex` badge, and keep their Codex thread title. A Source filter (All / Claude Code / Codex) is available in the web UI, over MCP, and on `recall list --source codex`. Codex tokens are priced from a separate OpenAI rate table, so cost across both agents lands in one set of views.

For Codex, Claude Recall reads exactly two paths: `~/.codex/sessions/**/rollout-*.jsonl` and `~/.codex/session_index.jsonl`. It never writes under `~/.codex`, and reading Codex adds no network activity of its own. That is a statement about reading, not an exemption: once a Codex session is indexed it is an ordinary row, so sharing a recap card, opt-in AI tagging and the Generate title / synopsis action reach it exactly as they reach a Claude Code session. None of them filters by agent, and none of them runs unless you switch it on or click it. The off switch is `recall codex off`, which writes `sources.codex.enabled` in `~/.recall/config.json`.

Beyond sessions, document buckets let you ingest your own Markdown and text files and search them from any repo, including from inside an agent session over MCP.

## Documentation

- [Quickstart](https://clauderecall.com/docs/quickstart): install + first search in two minutes
- [CLI reference](https://clauderecall.com/docs/cli): complete `recall` command surface
- [Web UI](https://clauderecall.com/docs/web-ui): browser-based session browser
- [VS Code extension](https://clauderecall.com/docs/vscode): inline session browsing in your editor
- [MCP server](https://clauderecall.com/mcp): expose Claude Recall to Claude Desktop, Cursor, and other MCP clients
- [Threads](https://clauderecall.com/docs/threads): how Claude Recall groups related sessions
- [Collections](https://clauderecall.com/docs/collections): curate and reuse session groups
- [Taxonomy](https://clauderecall.com/docs/taxonomy): tags, aliases, and notes
- [Semantic search](https://clauderecall.com/docs/semantic-vectors): local embedding-based search via bge-base-en-v1.5
- [Find sessions and terminals](https://clauderecall.com/docs/session-lookup): locate a past conversation by topic, id or an old alias, across Claude Code and Codex
- [Export](https://clauderecall.com/docs/export): pipe past sessions into a new agent run
- [Security and threat model](https://clauderecall.com/docs/security): local-first, no silent telemetry, daemon binds 127.0.0.1 only
- [FAQ](https://clauderecall.com/docs/faq): includes Codex CLI support, its exact read scope, and the off switch
- [Troubleshooting](https://clauderecall.com/docs/troubleshoot)

## Key concepts

- [Context Debt](https://clauderecall.com/blog/context-debt): the term. Cumulative loss of project memory across long-running AI coding sessions.
- [Where Claude Code stores sessions](https://clauderecall.com/blog/where-claude-code-stores-sessions): the JSONL transcripts at `~/.claude/projects/`
- [Pipe a past session into a new one](https://clauderecall.com/blog/pipe-past-claude-session-into-new-one): the canonical workflow
- [Local-first developer tools](https://clauderecall.com/blog/local-first-developer-tools-2026): the privacy and reliability case
- [Organizing thousands of AI coding sessions](https://clauderecall.com/blog/organizing-thousands-of-ai-coding-sessions): the operator playbook
- [Thirty terminals, twenty apps](https://clauderecall.com/blog/thirty-terminals-twenty-apps): the buyer's daily reality
- [Training the agent to think like you](https://clauderecall.com/blog/training-the-agent-to-think-like-you): why your past sessions are the best style guide
- [What MCP actually does](https://clauderecall.com/blog/what-mcp-actually-does): the protocol explained
- [Process tree terminal correlation](https://clauderecall.com/blog/process-tree-terminal-correlation): how Claude Recall ties terminals to sessions
- [Terminal scrollback vs IDE memory](https://clauderecall.com/blog/terminal-scrollback-vs-code-cursor-windsurf): why the terminal is the right surface
- [Claude Code best practices](https://clauderecall.com/blog/claude-code-best-practices)

## Comparisons

- [vs Cursor](https://clauderecall.com/vs/cursor): a terminal agent plus Recall vs Cursor's IDE-native AI
- [vs Obsidian](https://clauderecall.com/vs/obsidian): why Recall targets coding-agent transcripts, not general notes
- [vs CLAUDE.md files](https://clauderecall.com/vs/claude-md): why static memory files do not scale
- [vs claude-mem](https://clauderecall.com/vs/claude-mem): managed-memory vs local indexing
- [vs Aider](https://clauderecall.com/vs/aider): different tools, different mental models
- [vs Cline](https://clauderecall.com/vs/cline): terminal-native agents vs a multi-provider IDE extension
- [vs Windsurf](https://clauderecall.com/vs/windsurf): standalone IDE vs CLI plus extensions
- [vs llm CLI](https://clauderecall.com/vs/llm-cli): Simon Willison's llm vs purpose-built session memory
- [vs official memory MCP](https://clauderecall.com/vs/official-memory): Anthropic's reference memory server vs Recall

## Audiences

- [For engineers](https://clauderecall.com/for/engineers): the technical case
- [For vibe coders](https://clauderecall.com/for/vibe-coders): for builders without a CS degree

## Pricing and trial

- [Pricing](https://clauderecall.com/pricing): free tier, plus Pro at $49.69 one-time, lifetime. Lifetime updates. Never a subscription.
- [Install](https://clauderecall.com/install): npm install plus a 7-day Pro trial that auto-reverts to free
- [Refund policy](https://clauderecall.com/refund): 30-day no-questions

## Tools

- [Context Debt calculator](https://clauderecall.com/tools/context-debt-calculator): estimate your team's monthly context cost in dollars

## Optional

- [About](https://clauderecall.com/about): solo developer, no silent telemetry, no investors
- [Changelog](https://clauderecall.com/changelog): per-version release notes
- [Privacy](https://clauderecall.com/privacy)
- [Terms](https://clauderecall.com/terms)
- [llms-full.txt](https://clauderecall.com/llms-full.txt): the same data with key page content inlined