AgentsView tracks 60+ AI coding agents' cost locally

AgentsView is aimed at a specific side effect of AI-assisted coding: once a developer runs Claude Code, Codex, Cursor, Copilot or Gemini in parallel, their logs sit in separate places and there is no single view of how many tokens each one used or what that usage would cost at API rates. The tool addresses this by automatically collecting those local logs, so there is little setup work after installation, and by offering cross-agent full-text search, including working Japanese search.

Beyond usage charts and cost estimates, the dashboard shows when agents ran, how many ran at once, and active versus idle time, and an evaluation view scores tool-call failures, retries and context load. An experimental recall feature extracts reusable procedures and cautions from past conversations for the user to approve or hold. For teams, the MCP server lets an AI agent itself search past sessions, and histories from multiple PCs can be merged into PostgreSQL, with ClickHouse or DuckDB available for analysis.

The setup shown is Windows 11: a PowerShell install command followed by agentsview server and a browser at http://127.0.0.1:8080. Because everything runs locally, the data question shifts from which vendor holds the logs to the user's own machine and, optionally, their own PostgreSQL instance — something engineering leads overseeing several agents may weigh when deciding whether to adopt it.