recall skill
A skill for searching past sessions: a full-text index of Claude Code, Codex, pi and Grok chats in one SQLite database
- 145 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The skill reads your past sessions and stores their text in the local `~/.recall.db` database; it uses no network.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/arjunkmrm-recallDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skill reads your past sessions and stores their text in the local `~/.recall.db` database; it uses no network.
This is third-party code. Review the repository files before installing.
What it does
recall finds an old conversation with an agent by words from it. The `recall.py` script builds an SQLite FTS5 index in the `~/.recall.db` file over the sessions of four agents: Claude Code (`~/.claude/projects/`), Codex (`~/.codex/sessions/`), pi (`~/.pi/agent/sessions/`) and Grok (`~/.grok/sessions/`). Only user and assistant messages are indexed, while tool calls, thinking blocks and images are skipped. The first run indexes all sessions in a few seconds, and after that only new and modified files are updated, by modification time. Results are ranked by BM25 with a small recency boost: recent sessions get up to 20 percent, decaying with a 30-day half-life. English uses Porter stemming, while Chinese, Japanese and Korean use a separate trigram table; the query is routed automatically. A query accepts quoted phrases, `AND`, `OR`, `NOT` and prefixes like `buffer*`. The `--project`, `--days`, `--source` and `--limit` flags narrow the output. Without query text the script lists all sessions in the period. Each result carries an agent tag, a highlighted excerpt and a file path, and `read_session.py` reads the raw transcript.
Who it is for. Developers with hundreds of Claude Code and Codex sessions who need to recall which conversation discussed a function, an error or a decision.
Good fit when
- A week ago the agent solved an odd build error, and that conversation must be found to repeat the fix.
- A session was closed by accident and must be resumed with `claude --resume` or `codex resume` using the found identifier.
- A list of sessions from the last day is needed without text search, to pick the right one.
- Work happens in several agents, and the search has to cover Claude Code, Codex, pi and Grok at once.
- The history has messages in Chinese, Japanese or Korean, which ordinary word search skips.
Not a fit when
- Persistent memory that the agent injects into new sessions is needed: recall only searches and adds nothing to context.
- The history lives in another agent such as Cursor or Gemini CLI: there are four sources and the rest are not read.
- Search over tool calls or model thinking is needed: they do not enter the index.
Example request
Find a past session where we discussed the SQLite migration and show how to resume it.Limitations
The skill reads only four session folders: Claude Code, Codex, pi and Grok. The `~/.recall.db` index holds the text of your conversations in plain form, so access to the file should be restricted. Older versions need reindexing: when upgrading from 0.3.x, 0.4.x and 0.2.x run `--reindex` once to pull in pi, Grok and Chinese, Japanese and Korean text. The last commit dates from July 2026. There are no dependencies, only Python 3.9 or newer is needed.
How to disable. Delete the `recall` folder from the agent skills directory and the `~/.recall.db` file that holds the index. The skill does not modify the original sessions.
FAQ
How do I install recall in Claude Code?
Run `npx skills add arjunkmrm/recall` and restart the agent. After that the `/recall` command or a question like 'find a past session about foo' starts the search.
How do I resume a found session?
Go to the project folder and run `claude --resume SESSION_ID` for Claude Code or `codex resume SESSION_ID` for Codex. For pi and Grok the commands are `pi --session SESSION_ID` and `grok --resume SESSION_ID`.
What is indexed and where is the index stored?
User and assistant messages from the four session folders. The index sits in `~/.recall.db`, and tool calls and thinking blocks do not enter it.
Are keys or services needed?
No. The script runs on the Python 3.9+ standard library, and all search happens locally in SQLite with no text sent out.
How do I search for an exact phrase or a word combination?
Put the phrase in quotes: `"ACP protocol"`. For boolean queries use `rust AND async`, `tauri OR electron` and `NOT deprecated`, and `buffer*` for a prefix.
Discussion
No questions or reviews yet.
Related
SkillsTerminal search across Claude Code, Codex, OpenCode and other agent sessions, with a jump back into the right chat
Local memory for Claude Code: a single binary indexes past conversations and injects relevant context into new sessions.
About a hundred skills for Claude Code: meetings, research, publishing, TDD, Telegram, email and personal analytics
Local browsing, search and cost tracking for Claude Code, Codex and other agent sessions in one web UI
An MCP server and CLI that find in Claude Code and Codex sessions why a file became what it is: who asked the agent what, when and what changed
A Codex skill for safe local-state maintenance: a report-only default, backup and archive instead of deletion, log and worktree cleanup
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| recall (this card) | no | 145 | |
| Recall | no | 765 | a same-named plugin by another author: a local session log and project summary; this one searches finished sessions of four agents |
| Coding Agent Session Search (cass) | no | 1,200 | a separate CLI and TUI for searching the history of many agents; this is a dependency-free skill on Python and SQLite inside the agent |
README in short
The documentation opens with a question about a lost session and explains how the skill builds an index over four sources. Diagrams show the flow from session files to the database and the query route: detection of Chinese, Japanese and Korean text, search, ranking and the recency boost. Then it lists index properties and the command to run tests on `unittest`.