agent-sessions

Why we included this project

Most quota meters show one total percentage, so when several coding agents share your Codex or Claude limits you can't tell which session burned the hours. Agent Sessions attributes that burn to the individual session, and it can price each model at its own rate, which matters when an orchestrator spawns subagents and you want costs It also indexes the transcripts those agents already write to disk, so you can search across Codex, Claude Code, OpenCode, Cursor, Copilot CLI, and a dozen more from one place, browse images, and resume a session with one click where the CLI supports it. Everything stays on your Mac with no telemetry, a real plus if you treat session history as sensitive. For anyone juggling multiple AI coding tools, it does two useful things: shows which agent is eating your limits and surfaces work that would otherwise stay buried in scattered local logs.