Find the session you remember,
even when you don't remember its title.
Ranked full-text search across the coding-agent sessions already on your machine.
Claude Code, Codex, Cursor CLI, opencode and Gemini CLI.
agsearch indexes the local transcripts your coding agents already write. Search them in one ranked list, preview the matching lines, and resume the original session in the tool it came from. Everything stays on your machine.
brew install devcodes9/tap/agsearch
agsearchType anything you remember from a past conversation. Select a result to resume it.
Homebrew also installs fzf, which the interactive interface needs.
To run one search without installing anything:
uvx agsearch -n "stripe tax id"And inside Claude Code, so Claude searches your past sessions itself rather than telling you it has no record of them:
/plugin marketplace add devcodes9/agsearch
/plugin install agsearch@agsearch
- Full-conversation search. Search user prompts and assistant replies, not only titles and session metadata.
- One list for every tool. Sessions from all five agents appear together, each row named after the tool it came from. Adding another agent is a parser plus one entry in the source table, with no change to search or ranking.
- Ranked results. BM25 ranking favors focused sessions and shows matching lines in context.
- Preview, read, or resume. Inspect a match, open the transcript in a pager, or return to the original session.
- Your agent can search too. A Claude Code skill, so Claude finds the earlier conversation itself instead of answering that it has no record of it.
- Fully local. No uploads, API keys, hosted index, or network calls.
- Fast warm searches. A per-file cache reparses only transcripts that changed.
Recommended because it installs both agsearch and the fzf dependency:
brew install devcodes9/tap/agsearchuv tool install agsearch
# or
pipx install agsearchThe interactive interface needs fzf 0.35 or
newer — that is the release which added the start event agsearch binds. Some distributions
package an older one; fzf's own install script is the fallback. Without fzf,
agsearch -n "query" still prints ranked sessions.
curl -fsSL https://raw.githubusercontent.com/devcodes9/agsearch/main/install.sh | shThis installs the latest release to ~/.local/bin. Set PREFIX to change the destination or
AGSEARCH_VERSION to pin a release.
agsearch requires Python 3.9 or newer and has no Python package dependencies.
brew update && brew upgrade agsearch # Homebrew
uv tool upgrade agsearch # uv
pipx upgrade agsearch # pipx
pip install -U agsearch # pipRe-run install.sh to update an install-script copy; it overwrites in place. /plugin updates
the Claude Code skill.
The first search after an update can rebuild the index, which takes a few seconds on a large corpus. That happens when a release changes how transcripts are parsed, and only once.
agsearch # browse all sessions in the interactive interface
agsearch "stripe tax id" # open with an initial query
agsearch -n "stripe tax id" # print ranked sessions as plain text, no fzf
agsearch read <session-id> # print a whole session, without resuming it
agsearch --here "webhook" # search only the current project
agsearch -p myapp "migration" # search projects whose path contains "myapp"
agsearch --thinking "query" # include assistant thinking blocks
agsearch --no-resume "query" # print the selected resume command
agsearch --reindex # rebuild the transcript cache
agsearch --version # print the installed version-n prints one entry per session as plain text, led by the session id, and drops colour
whenever it is not writing to a terminal. That makes the search loop scriptable:
agsearch -n "webhook retry backoff" # ranked sessions, one entry each
agsearch read 3f2a1c4e-... # the whole conversation, no resume, no tokensRanking is the same as the interactive list, so a term you half-remember or mistype finds
the same session either way. Piped, the session ids shorten to a unique prefix, the columns
lose their padding, and read prints the start and end of a long session rather than all of
it. A terminal sees none of that.
agsearch ships a Claude Code skill. Install it from inside Claude Code:
/plugin marketplace add devcodes9/agsearch
/plugin install agsearch@agsearch
From the next session on, Claude searches your transcripts itself when you refer to work from an earlier conversation:
you: what did we decide about the webhook retry backoff?
Claude: runs
agsearch -n "webhook retry backoff", reads the top hit, answers from it
It also covers handoff, which resuming cannot do. claude --resume moves you back into the old
session in its own directory; the skill carries that session's context forward into the one you
are in now, so you can pick the work up in a different repository or on a different branch.
The skill is a single markdown file. Read it before installing. If you would rather not add a marketplace, copy it instead:
mkdir -p ~/.claude/skills && cp -r skills/agsearch ~/.claude/skills/Either way it needs the agsearch binary, which the installation section above covers.
| Key | Action |
|---|---|
| Enter | Resume the selected session |
| Ctrl-O | Read the full conversation in your pager |
| Ctrl-Y | Copy the resume command |
| Ctrl-/ | Toggle the preview pane |
Selecting a result resumes the session in the tool that created it, from that session's project directory. The current query is copied to the clipboard so you can find the same text after resuming.
For a global shortcut, see the hotkey guide.
Claude Code's /resume picker and codex resume are good when you remember a session's title,
branch, directory, or first prompt. They search metadata about the session.
agsearch searches the conversation itself. It also combines both tools in one list and includes
Claude Code SDK and -p sessions that do not appear in the native picker.
Use the native picker when you remember what the session was called. Use agsearch when you remember what was said.
agsearch drops common stopwords, applies conservative stemming, and ranks matching sessions with
BM25 across three weighted fields: title and project, first prompt, and full transcript. Sessions
covering more query terms rank first; relevance, recency, and previous resumes break close ties.
Rare long typos fall back to subsequence matching, so conection pool still finds the
session about connection pools.
Search is lexical, not semantic. It will not match concepts expressed with completely different words, and the first result is not guaranteed to be the session you intended.
agsearch reads:
| Agent | Read from | Resumed with |
|---|---|---|
| Claude Code | ~/.claude/projects/**/*.jsonl |
claude --resume <id> |
| Codex | ~/.codex/sessions/**/*.jsonl |
codex resume <id> |
| cursor-cli | ~/.cursor/projects/**/agent-transcripts/ |
cursor-agent --resume <id> |
| opencode | ~/.local/share/opencode/opencode.db |
opencode --session <id> |
| Gemini CLI | ~/.gemini/tmp/**/chats/*.json |
gemini --session-file <path> |
opencode keeps sessions in SQLite; agsearch opens it read-only and reads message records
only. Gemini's --resume takes a project-scoped index number rather than a stable id, so
resume goes through the transcript file instead.
Cursor and Gemini are read from their CLI's storage. Chats made in the Cursor IDE are kept elsewhere and are not indexed, which is why the column names the CLI.
Its cache lives under ~/.cache/agsearch/. Transcript parsing and ranking happen locally, and
only changed files are reparsed.
Important
Claude Code deletes transcripts after 30 days by default. To keep a longer searchable history,
set cleanupPeriodDays in ~/.claude/settings.json:
{ "cleanupPeriodDays": 365 }agsearch never changes this setting.
- Claude Code subagent transcripts are folded into their resumable parent session.
- SDK and other automated sessions remain searchable but rank below user-started sessions.
- Sessions from deleted worktrees resume from the nearest existing parent directory.
- Recently active sessions are marked
●and require confirmation before reattaching. - Forked Claude Code sessions are marked
fork, and name the branch they split from.
git clone https://github.com/devcodes9/agsearch.git
cd agsearch
python3 -m unittest discover -s testsChanges to ranking should include a regression case in tests/. See the
changelog and
open issues.