# AgentsView
**AgentsView is a local-first analytics dashboard that pulls session data from 40+ AI coding agents into one place to search transcripts, track tokens, and see what you actually spent.** One binary, no accounts, everything local. Go-based, MIT.
Where [[abtop]] is the live `htop` view of running agents, AgentsView is the historical analytics layer: it reads the transcripts your agents already wrote to disk and turns them into searchable, costed history.
## What it gives you
- **Full-text search** across every message via SQLite FTS5.
- **Cost tracking** using LiteLLM pricing, prompt-cache-aware (separate input, output, cache-creation, cache-read).
- **Dashboards.** Activity heatmaps, tool usage, velocity, per-model breakdowns.
- **Live updates** via server-sent events for active sessions.
- **Multi-backend.** SQLite primary, PostgreSQL for team sync, DuckDB for portable mirrors.
- **Export.** HTML sessions or GitHub Gist publishing.
It auto-detects 40+ agents (Claude Code, Codex, Cursor, Forge, DeepSeek, OpenHands, Qwen, ...) by scanning their session directories. Install via curl/PowerShell one-liner, Homebrew cask, or Docker; there's also a Tauri desktop wrapper.
## Custom model pricing
Newly released models often ship before AgentsView knows their price. You add it yourself in `~/.agentsview/config.toml`. Per Simon Willison's TIL, adding [[Claude Fable 5]] looked like:
```toml
[custom_model_pricing."claude-fable-5"]
input = 10.0
output = 50.0
cache_creation = 12.50
cache_read = 1
```
He ran it via `uvx agentsview usage daily` for a terminal table and `uvx agentsview serve` for the web dashboard (port 8080). The example makes the point that Fable 5 costs about double Opus per token.
## References
- https://www.agentsview.io/
- https://github.com/kenn-io/agentsview
- https://til.simonwillison.net/llms/agentsview-custom-model-price
## Related
- [[abtop]]
- [[Claude Code]]
- [[Codex CLI]]
- [[Cursor.com]]
- [[LiteLLM]]
- [[Claude Fable 5]]
- [[AI Agents]]