AgentsView

A working developer now runs several AI coding agents at once — Claude Code in one terminal, Codex in another, maybe Cursor, Gemini CLI, or Copilot on top. Each quietly writes its own session and transcript files somewhere on disk, in its own format, and none of them talk to each other. AgentsView, from Wes McKinney (the creator of pandas), auto-discovers those scattered files for 40-odd agents, loads them into a local SQLite index, and puts one search box, one dashboard, and one cost ledger over the whole pile. It is an MIT-licensed single binary you install with one command and point at your own machine; the data never leaves it.

The reasoning behind a change you made in October stays findable — whichever of a dozen agents you happened to be driving that day.

The part that makes it more than a log viewer is the accounting. Existing cost trackers like ccusage read one agent and re-parse the raw files on every run; AgentsView keeps everything pre-indexed, so its `usage` command is a drop-in replacement that answers the same questions across all your agents at once. It is how Simon Willison discovered, after shipping an AI-written release of his sqlite-utils toolkit, exactly what the agent's tokens had cost — reconstructed from transcripts already sitting on his laptop, with a single command run inside the session he was measuring.

The obvious use is a monthly spend audit; the quieter one is memory. Full-text and optional semantic search across months of sessions means the reasoning behind a change you made in October is findable, whichever agent you happened to be driving that day. The token-cost half is the newest and, by McKinney's own admission, still rough — parsers for some agents only recently learned to emit token counts, and the eye-catching speed figures come from AgentsView's own benchmark. But the core is a real, inspectable tool, and it is a neat sign of the times: it takes a dedicated indexer just to keep track of how many different robots are writing your code.