name	recall
description	Search the user's own past AI chat history across Cursor, Claude Code, Codex, and pi via the `recall` CLI. Use when the user references earlier work — "how did we fix…", "what did I decide about…", "continue the…", "didn't we already…" — that may live in a prior conversation.

recall — search past AI chat history

Read-only full-text search over your past Cursor / Claude Code / Codex / pi sessions. Use it whenever the user points at past work not in current context.

If recall is missing or the index is stale, run recall index (incremental, ~ms once built). To install: go install github.com/pratikgajjar/recall@latest.

Workflow: find, then read

recall find "import cycle proto" --limit 5   # ranked hits; --json to parse
recall show cursor:94dc8775-…                # full transcript of a hit
recall last --repo .                          # most recent session in THIS repo

Each hit gives a session_id and a msg index at the matched message.

Big sessions: slice, don't dump

Sessions can be 30k+ messages. recall show <id> takes:

--outline                 # [N] role: first-line — a table of contents
--range 305:315           # window (also :100, -50:); indices are absolute
--role user,assistant     # drop tool noise (often ~50% of an agent loop)

After a hit at msg=N, prefer recall show <id> --range N-5:N+5. For an unfamiliar session, start with --outline --role user — the prompts are its spine.

Tags: durable bookmarks (git-tag style)

Tags survive recall index --full (stored apart from the disposable index).

recall tag cursor:94dc8775-… deploy-rca   # remember this session
recall tag                                # list all tags + counts
recall tag -d <id> <tag>                  # remove   |  -l [id]  list
recall sessions --tag deploy-rca          # filter by tag

--tag is the one filter selector (repeatable, AND). source is a reserved facet on it — filter with --tag source:cursor (no separate --source flag); you can't author a facet as a tag.

Flags & tips

--repo PATH (. = cwd) · --after 7d / --before WHEN (alias --since) · --limit N · --json

find, sessions, and show print terse next:/prev: page commands at the bottom — follow them to traverse everything. Across sessions pages by time (--before/--after); within show by message window (--range). WHEN is a duration (7d), an epoch-ms (paste started_at_ms from --json), or YYYY-MM-DD.

Query with concrete identifiers (errors, symbols, paths) — it's FTS, not semantic.
recall related <id> widens to neighbouring sessions on the same topic.
Covers only your own AI chats — not web/docs. --full only if the index looks corrupt.

recall — search past AI chat history

Read-only full-text search over your past Cursor / Claude Code / Codex / pi sessions. Use it whenever the user points at past work not in current context.

If recall is missing or the index is stale, run recall index (incremental, ~ms once built). To install: go install github.com/pratikgajjar/recall@latest.

Workflow: find, then read

recall find "import cycle proto" --limit 5 # ranked hits; --json to parse recall show cursor:94dc8775-… # full transcript of a hit recall last --repo . # most recent session in THIS repo

Each hit gives a session_id and a msg index at the matched message.

Big sessions: slice, don't dump

Sessions can be 30k+ messages. recall show <id> takes:

--outline # [N] role: first-line — a table of contents --range 305:315 # window (also :100, -50:); indices are absolute --role user,assistant # drop tool noise (often ~50% of an agent loop)

After a hit at msg=N, prefer recall show <id> --range N-5:N+5. For an unfamiliar session, start with --outline --role user — the prompts are its spine.

Tags: durable bookmarks (git-tag style)

Tags survive recall index --full (stored apart from the disposable index).

recall tag cursor:94dc8775-… deploy-rca # remember this session recall tag # list all tags + counts recall tag -d <id> <tag> # remove | -l [id] list recall sessions --tag deploy-rca # filter by tag

--tag is the one filter selector (repeatable, AND). source is a reserved facet on it — filter with --tag source:cursor (no separate --source flag); you can't author a facet as a tag.

Flags & tips

--repo PATH (. = cwd) · --after 7d / --before WHEN (alias --since) · --limit N · --json

Query with concrete identifiers (errors, symbols, paths) — it's FTS, not semantic.

recall related <id> widens to neighbouring sessions on the same topic.

Covers only your own AI chats — not web/docs. --full only if the index looks corrupt.