agtailCross-agent search for coding-agent histories

Forensic, after-the-fact search across Claude Code and Codex — "where did that operation happen, and what did it do?" Search first; viewing is secondary.

Get started

Concepts

Try the playground

Search-first

grep is a first-class citizen. Cross-agent search built around compound filters — tool × model × cwd × time range × agent × origin. Timelines and stats are byproducts.

Agent-agnostic

Claude Code and Codex are normalized into one schema and read through the same interface. One entry point, however many agents you add.

Read-only

Never edits. It faithfully replays and extracts the fixed past. The original transcripts stay put; agtail owns the index and the search.

CLI and Web UI

Query from the terminal with agtail grep / list / show / stats, or browse a search-first two-pane UI with agtail serve.

Tokens & cost

Per-turn token counts and approximate cost. Prices are sourced from LiteLLM — a model it doesn't list shows "unknown" rather than a guess.

Hooks & plugin attribution

Hook firings are first-class, searchable events — which event fired, which tool triggered it, the text it injected, and which installed plugin owns it.

Programmatic & subagent sessions

SDK-spawned and subagent sessions are detected and nested under context. A spawned review is attributed back to the plugin that launched it.

Cross-machine sync

Export sessions to a portable bundle and import them elsewhere into named collections, so one viewer can audit several machines or people side by side.