The receipt printer for your AI coding habit

What it does

agentsview slurps session history from Claude Code, Codex, Cursor, and 20+ other agents into a local SQLite database, then gives you a web UI to search, browse, and tally costs. It also replaces ccusage with a CLI that queries that same index—no re-parsing raw logs on every run.

The interesting bit

The project treats agent sprawl as a data-integration problem, not a dashboard problem. It auto-discovers 25+ agents from their default directories, handles encrypted transcripts (Antigravity CLI needs a sidecar helper), and even pushes to PostgreSQL for team-wide read-only dashboards. The “100x faster than ccusage” claim comes from indexing once and querying SQLite rather than re-scanning session files.

Key highlights

• One static binary, no accounts, runs fully offline
• FTS5 search across all message content; live SSE updates as sessions grow
• Cost tracking with LiteLLM pricing, prompt-cache awareness, and per-model breakdowns
• Keyboard-first web UI (j/k, Cmd+K, ? for shortcuts)
• Optional PostgreSQL sync for team analytics; Docker image included
• Exports sessions as HTML or GitHub Gist

Caveats

• Docker setups require explicit read-only mounts per agent directory; miss one and that agent vanishes from the UI
• Remote access (SSH tunnels, Codespaces, etc.) needs --public-url set exactly right or API calls 403
• Git-derived outcome metrics (commits, LOC changed) are opt-in because they “can be slow or brittle on large/missing repos”
• Antigravity CLI full transcripts need a separate agy-reader tool running alongside

Verdict

Worth a look if you’re running multiple agents and want to know where your API budget went—or need to grep your own prompts later. Skip it if you only use one agent and already have ccusage working fine; the multi-agent aggregation is the whole point.