# Search

Full-text search across every agent session with ai-hist search and recent, using entry IDs as the handle to drill deeper.

Rendered page: https://agentrelay.com/docs/loop/search
Markdown endpoint: https://agentrelay.com/docs/loop/markdown/search.md

---

Search is the Recall lens. Once you have [synced](/docs/loop/sync), every prompt you and your team have ever sent to an AI agent is one query away — across Claude Code, Codex, Cursor, Grok, OpenCode, Agent Relay, and trajectories at the same time. Search is powered by SQLite FTS5, so it is fast even over tens of thousands of entries.

## Full-text search

Search across all sources with a query string:

```bash
ai-hist search "authentication bug"
```

Results come back ranked, each with an entry ID in the form `#NNN`. That ID is your handle to drill in further.

## Filters

Narrow a search by source, project, or result count:

```bash
ai-hist search "refactor" --source claude --limit 10
ai-hist search "deploy" --source relay
ai-hist search "retry policy" --source trajectory
ai-hist search "deploy" --project relay
```

- `--source` — restrict to one source. Valid values are `claude`, `codex`, `cursor`, `grok`, `relay`, `trajectory`, and `opencode`. See [Sources](/docs/loop/sources).
- `--project` — restrict to a project (matched against the recorded project path).
- `--limit` — cap the number of results returned.

## Recent prompts

To see what you have been working on most recently without a query, use `recent`:

```bash
ai-hist recent                              # last 20
ai-hist recent 50                           # last 50
ai-hist recent --source claude --project my-app
```

`recent` accepts the same `--source` and `--project` filters as search.

## Entry IDs are the drill-in handle

Search and recent both surface entry IDs (`#NNN`). Use them to move from a hit to the full picture:

```bash
ai-hist search "deploy"   # find #4521 in the results
ai-hist show 4521         # full prompt, session info, resume command
ai-hist context 4521      # what else was happening in that session, plus nearby entries
ai-hist session <id>      # browse the full conversation
```

This is the core Recall loop: search broad, find the entry, then [drill into the session](/docs/loop/sessions) and resume it.

> Searching for a problem you have hit before is the fastest way to recover a working approach — find the prompt that solved it last time, then reuse or resume it.

- [Sessions](https://agentrelay.com/docs/loop/sessions): Show, context, session, and per-tool resume commands.
  - [Sources](https://agentrelay.com/docs/loop/sources): The full list of sources and `--source` filter values.