AgentsView — Local Agent Session Browser by Wes McKinney

URL: https://www.threads.com/@ryanchou0210/post/DZbipuUEXtk
Date Saved: 2026-06-11
Source: Threads (@ryanchou0210)
Tags: ai-tools, ai-engineering

Summary

Ryan Chou 分享了 AgentsView (agentsview.io)，由 Pandas 作者 Wes McKinney 開發的本地工具，用來瀏覽和搜尋所有 coding agent 的對話歷史和花費統計。

What is AgentsView?

A local-first, open-source tool (github: wesm/agentsview) that reads session files left by coding agents on your machine and provides a web UI to browse, full-text search, and get statistics on all past agent conversations.

Key Features:

Supports 20+ agents: Claude Code, Codex, Cursor, Copilot, Gemini, etc.
Auto-scans the correct directories — zero config
Local web UI at 127.0.0.1:8080
SQLite-backed (stored at ~/.agentsview/)
No account, no cloud, localhost-only by default (only outbound = update check, can be disabled)

Usage Tracking (killer feature):

agentsview usage daily command — like ccusage but faster
On 20K+ sessions: 80–220x faster than npx ccusage
Aggregates token spend across ALL agents (not just Claude Code)
Separates prompt cache read/write costs
Pricing via LiteLLM rates with offline fallback
Statusline showing daily spend
Custom pricing in ~/.agentsview/config.toml for new models

Simon Willison’s example:

Claude Fable 5 wasn’t in pricing DB yet
Added custom pricing (input $10/M tokens, output $50/M — ~2x Opus)
Found he burned $82 worth of tokens in 4.5 hours (covered by $100/mo Claude Max)

Other features:

Browse prompts, thinking process, tool calls per conversation
Full-text search across months of history (“how did I fix that bug?”)
Can generate daily summaries via Claude/Codex
Install: uvx agentsview or desktop app from agentsview.io