
kenn-io/agentsview: Go CLI + dashboard that indexes 27 agents into SQLite, tracks costs 80–220× faster than ccusage. v0.33.0 out today.
kenn-io/agentsview — a local-first Go CLI + web dashboard that auto-discovers and indexes sessions from 27 coding agents into a SQLite database, then gives you full-text search, per-session cost breakdowns, and usage analytics. ~2,000 GitHub stars in roughly 10 hours. MIT license. 1Note on categorization: agentsview is not a traditional Agent Skill — it has noSKILL.md, is not listed on skills.sh, and is not installed vianpx skills add. It's a standalone binary that sits alongside your agents and observes them. That's the correct frame for evaluating it.
| Field | Value |
|---|---|
| Repo | kenn-io/agentsview |
| Author | Wes McKinney (creator of pandas, founder of kenn-software, GP at Composed Ventures) |
| License | MIT |
| Latest version | v0.33.0 (Jun 12, 2026) |
| Stars / forks | ~2,000 / 191 1 |
| Supported agents | 27 (Claude Code, Codex, Gemini CLI, Cursor, OpenCode, OpenHands, Amp, Kimi, Qwen Code, Forge, Warp, Zed, and more) |
| Backend options | SQLite (default), PostgreSQL (team push), DuckDB (v0.33.0 mirror) |
| Runtime | Go 1.26+ with CGO; pre-built binaries available |
agentsview serve scans your machine, finds every agent session it recognizes across the 27 supported tools, and syncs them into a local SQLite database with FTS5 full-text indexing (SQLite's built-in search engine). No config file, no API key, no setup wizard. As Wes McKinney described the design intent: 2"A new pattern is emerging for apps that 'just work' with your agents. AgentsView just used my local Claude Code and I didn't even have to set anything up."

prod_datasette_agent) had consumed $74.06 on one day, with cache savings of $516.62 offset against that. 3agentsview sync: sync to a remote rather than only localcurl -fsSL https://agentsview.io/install.sh | bashbrew install --cask agentsviewpip install agentsview
# or, without permanent install:
uvx agentsview servedocker run -p 8080:8080 ghcr.io/kenn-io/agentsview:latest.dmg (macOS), .exe / ARM64 (Windows), or .AppImage (Linux) from GitHub Releases. The official site describes the desktop app as the recommended path. 5agentsview serve
# opens http://127.0.0.1:8080 — the server binds to localhost only by defaultagentsview usage daily
# shows last 30 days of per-day cost totals
agentsview usage daily --breakdown
# breaks each day down by model
agentsview usage daily --since 2026-05-04 --agent claude
# filter to a specific agent and date range"Run 'uvx agentsview usage daily --since 2026-05-04' to see how much that would have cost you in API costs (the price Uber pay) as opposed to just your individual subscription costs." 3
agentsview usage statuslineagentsview stats
# prints: total sessions, prototype distribution, average duration,
# peak context size, cache economics, hourly activity breakdownagentsview session usage <session-id>
# token counts + cost estimate for one sessionagentsview pg pushagentsview duckdb push # mirror to a local DuckDB file
agentsview duckdb serve # start a Quack endpoint for remote access"quite a nice tool for understanding agent usage and trends, you can plot by project/model/harness. then you can also do funny trend analysis for model tic words."

"agentsview caught my eye today. I like that it indexes local agent sessions into SQLite, then gives a usage UI plus a 100x faster ccusage replacement across Claude Code, Codex, and 20+ agents. Feels inevitable."
ccusage is one of the concrete selling points in the README: on a 22,000-session local database, agentsview usage daily runs 80–220× faster than npx ccusage@latest daily, because agentsview queries a pre-indexed SQLite database rather than re-parsing raw JSONL on every call. 1"Agentsview is a local-first, open-source way to view agent activity across your coding tools, including Claude Code, Codex, and now OpenHands. That makes it much easier to search, review, and summarize what your agents are doing."
"一个工具解决了 AI 编程的一个隐形痛点" — a tool that addresses the invisible pain point of AI coding: knowing which session burned the most tokens, which task the agent handled worst, and what kept requiring rework. 9
ccusage covers the cost-tracking need with less setup overhead.npx skills add. It isn't. There's no SKILL.md. It won't intercept your agent's behavior or add guardrails. It only observes.agentsview stats will tell you within seconds of first launch.ccusage and find it slow on large session archives. The 80–220× speedup claim is based on a 22,000-session dataset — at that scale, the SQLite pre-indexing approach has a clear practical advantage.# Quickest no-install test:
uvx agentsview serve
# macOS permanent install:
brew install --cask agentsview
# Cross-platform:
curl -fsSL https://agentsview.io/install.sh | bash
# Start the dashboard:
agentsview serve
# → http://127.0.0.1:8080
# Daily cost check:
agentsview usage daily --breakdowngithub.com
agentsview.io
x.com
x.com
x.com
x.com
github.com
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.