↓Skip to main content
  1. Tools/
  2. AI Sessions/

AgentsView

·2 mins·
ai-ready AI Sessions external Ai-Sessions Claude-Code Analytics Search Sqlite Sessions Multi-Agent Mcp
Language: Go
License: MIT
Maturity: stable
Verified: 2026-08
Table of Contents

AgentsView is a local-first platform that discovers and indexes AI coding session artifacts from 40+ agents – Claude Code, GitHub Copilot, Cursor, Codex, Devin, OpenHands, Windsurf, and many others – into a unified searchable interface. All data stays on the machine; the server binds to 127.0.0.1 by default.

What It Captures #

AgentsView reads agent session directories automatically at startup and indexes:

  • Conversation transcripts and message content
  • Token consumption (input / output / cache) and cost
  • Tool usage patterns and model selections
  • Timestamps, session duration, and project path

It does not require any configuration of the individual agents – it reads their existing session storage locations.

Storage #

Sessions are indexed into a local SQLite database with FTS5 full-text search. Optional mirrors include DuckDB files for analytical queries and PostgreSQL for shared team dashboards. The source session files (e.g. Claude Code’s JSONL in ~/.claude/projects/) are read-only; AgentsView does not modify them.

Search and Analysis #

  • Full-text search across all session content, any agent
  • Semantic search (vector embeddings, self-hosted)
  • Token and cost analytics by agent, model, project, or time range
  • “Recent Edits” feed linking sessions to the files they touched
  • Session export to HTML or GitHub Gist

MCP Server #

AgentsView exposes an MCP server so AI agents can query session history programmatically – enabling workflows where an agent reviews what a previous session did before continuing work.

Relevance to Archival #

AgentsView is primarily an analytics and search tool rather than an archival system in the preservation sense – it does not version the session data or provide immutable storage. Its value for the con/serve mission is as a discovery and access layer: it makes session artifacts findable across agents and queryable in aggregate, which is a prerequisite for deciding what is worth archiving into a DataLad dataset.

The session files AgentsView indexes – Claude Code JSONL under ~/.claude/projects/, Cursor logs under ~/.cursor/, etc. – are the actual preservation targets. Tools like ccexport and cctrace handle extracting and converting those files for archival; AgentsView handles finding and understanding them at scale.

See Also #

  • cctrace – lightweight Claude Code session capture
  • ccexport – converts Claude Code JSONL to markdown/JSON for archival
  • entire-io – git-native session archival with shadow branches
  • roborev – continuous code review database for AI-generated commits