agentsview lets you run a local web UI and CLI that indexes every AI coding agent session on your machine, stores them in SQLite (or DuckDB/PostgreSQL), and shows you token counts, costs, and usage patterns. You install it with a single script (curl https://agentsview.io/install.sh | bash), via Homebrew (`brew install --cask agentsview`), a Windows PowerShell command, or by pulling the Docker image (`ghcr.io/kenn-io/agentsview:latest`). On first launch it discovers sessions for Claude, Forge, Codex, OpenCode, etc., syncs them into a `~/.agentsview` data directory, and spins up a server at http://127.0.0.1:8080. You can daemonize it (`agentsview serve --background`), check status (`serve status`), or shut it down (`serve stop`). To expose beyond localhost—say through SSH port-forwarding or Codespaces—you restart with `--public-url` and, if needed, `--require-auth`. The Docker examples cover everything from a simple container to production setups. A `docker-compose.prod.yaml` mounts session directories read-only, persists data in a named volume, and binds to loopback by default. If you want PostgreSQL instead of SQLite, set `PG_SERVE=1` and point `AGENTSVIEW_PG_URL` at your database. For high-performance analytics, you can push your SQLite archive into a DuckDB mirror inside the container, then serve it read-only. There’s even a Quack example that exposes the DuckDB mirror over an authenticated HTTP API for remote dashboards. All mounts must be explicit—if you omit a directory for an agent, its sessions won’t show up. On the CLI side, `agentsview usage` replaces tools like ccusage. It gives sub-second summaries of token consumption and cost by model, date range, or agent (`daily`, `--breakdown`, `--since`, `--agent claude`). `agentsview session usage <id>` drills into a single session’s output tokens, peak context, and USD cost. JSON flags (`--json` or `--format json`) let you script these outputs. For higher-level analytics, `agentsview stats` outputs totals, session archetypes (quick, deep, marathon), distributions for duration or peak context, and even Git-based outcomes (commits, PRs) if you opt in. Everything runs locally, so there’s no external account or cloud dependency—just your data, your logs, your costs.