AgentsView — Browse and Search Past AI Coding Sessions

https://www.agentsview.io/

Why This Matters

This one is directly adjacent to our claude-code-sessions and claude-code-tool-traces Chroma collections — it's a packaged, local-first tool for exactly the problem we've been ingesting into the corpus ourselves. Browse, search, and analyze your past AI coding sessions across 20 different agent tools, with full-text search, token/cost tracking, AI-generated summaries, and activity visualizations.
The repo is Wes McKinney's (creator of pandas, Apache Arrow), which is a meaningful signal about the engineering bar. MIT-licensed, version 0.29.0 just hit Homebrew cask.

What It Actually Does

  • 20 agent integrations, auto-discovered. Claude Code, Codex, Copilot, Cursor, Gemini, OpenHands, VS Code Copilot, Positron Assistant, and more — it finds the session directories on disk without configuration.
  • Imports from chat too. Pulls Claude.ai and ChatGPT chat history alongside the agent CLI sessions.
  • Local-first by default. SQLite + embedded web frontend, no cloud, no accounts. Optional PostgreSQL sync for multi-machine or team setups — opt-in only.
  • Performance claim worth noting. The agentsview usage CLI is stated as 80–220× faster than comparable token/cost analyzers on large histories. We have large histories.
  • AI-generated session summaries. Useful for reconstructing "what was I doing in that 4-hour Claude Code run last Tuesday."

Where It Fits in Our Workflow

We already have the four-collection corpus (context-vigilance-corpus, lossless-changelog, claude-code-sessions, claude-code-tool-traces) wired into Chroma for the search-first discipline. AgentsView is the UI counterpart: when the question is "let me eyeball a specific session" rather than "what did we decide about X," the local web app is the right surface.
The 20-agent breadth also matters for the Lossless tree — we use Claude Code as the primary harness, but Trae, Codex, and Cursor have shown up in side projects. A single browser across all of them is a real win versus per-tool session directories.

Two Install Paths

bash 
# Cask (recommended for the desktop app)
brew install --cask agentsview

# Or pip/uvx if you only want the CLI
uvx agentsview usage
Source: https://github.com/wesm/agentsview

Open Question

Whether the AgentsView SQLite schema is rich enough that we could point our Chroma ingestion at it instead of parsing raw Claude Code session JSON ourselves. If so, that simplifies ai-labs/context-vigilance-kit/scripts/ materially.