Local-first agent tracking
VibeLog records every agent session — prompts, tool calls, output, cost — to your own disk, and makes it readable. A flight recorder for the work you delegate.
Every session renders as a tape. After a day with VibeLog you stop reading transcripts first — you glance at the strip and know which of your agents needs you.
Steady rhythm — read, edit, test, commit. You can stop watching this one.
The long silence after minute one is a hung test suite. Worth interrupting.
Retries piling into red at the end. The transcript says why.
Mission control
Live sessions with their tapes, the queue behind them, and today's spend — the whole fleet at a glance. When something needs you, it's red; when nothing does, the page is quiet.
Transcripts
Every prompt, every tool call with its wall-clock time, every line the model wrote back. When a run goes sideways at 2am, the answer is in the log — not in your scrollback.
Analytics
Spend per day, per model, per session. Tool latency at p50 and p95. Costs are computed from recorded token counts at rates you control — so the number is yours, not a bill you find out about later.
Recordings live in ~/.vibelog on the machine that ran the agent. VibeLog works on a plane, in an air-gapped lab, and after we're gone.
One JSON snapshot in ~/.vibelog, fed from the transcripts your agent already writes. Parse it with a shell one-liner; grep it like anything else you own.
The workspace binds to localhost and makes no outbound requests. The only person who knows what your agents did is you.
Stop scrolling terminal history. Start keeping records.
Open mission controlfree while in beta · no account · npm install -g vibelogapp