Most teams stitch together per-agent logs, slow parsers, or online dashboards to answer simple questions like “which sessions spent the most tokens” or “how much did a model cost last week.” agentsview flips that workflow: it indexes agent session roots into a local SQLite archive and exposes fast, queryable analytics and a web UI so cost, token, and activity metrics are available immediately on your machine. Because ingestion is local-first and the server binds to loopback by default, it keeps session content off the network while still providing exports and optional shared backends.
What Sets It Apart
- Local-first, single-binary UX: run a CLI or a Tauri desktop app; no account or outbound indexing required. This makes it practical for developers who prefer offline tooling or who handle sensitive code/prompts.
- Token-aware cost accounting: integrates LiteLLM pricing (with offline fallback), is prompt-cache aware, and reports per-session output tokens, peak context, and USD estimates where pricing is known — faster than re-parsing raw session files on every run because everything is indexed in SQLite.
- Multi-agent discovery & cross-agent dashboards: auto-discovers sessions for Claude Code, Codex, Copilot CLI, Forge, OpenCode, Gemini CLI, Cursor, and many others, plus a per-model breakdown and activity heatmaps. Supports live updates via SSE and exports (HTML/Gist).
- Production-friendly sync options: push local data to PostgreSQL for team dashboards or mirror into DuckDB and expose via Quack; includes Docker/compose examples and a background pg push daemon for continuous sync.
Who It's For and Tradeoffs
Great fit if you run multiple local coding agents and need fast, private analytics: developers tracking spend across experiments, SREs diagnosing agent-driven automation, or teams building shared dashboards from local session archives. It’s also useful as a 100x-faster replacement for token-usage scripts like ccusage. Look elsewhere if your sessions live only in hosted/closed cloud accounts (no local session files), if you need SaaS multi-tenant sharing out of the box, or if you prefer a managed analytics product. Additional tradeoffs: discovery requires agent session directories (or mounting them into containers), the tool expects a modern Go build chain for local development, and some advanced features (Postgres sync, DuckDB mirror, Quack) add operational complexity.
Optional notes: telemetry is minimal and opt-outable (env var), the server defaults to loopback with Host-header validation to mitigate DNS-rebinding, and the project provides a desktop build and Docker images for quick local deployment. As of initial repository creation it has community traction and active features around token accounting, but adoption depends on whether your workflows generate discoverable local session files.
