PRODUCT · COST & OBSERVABILITY

See what every agent did, and what it cost.

AI agents are usually a black box with a surprise bill. Knitch shows you both. Every run is traced node by node, and every node carries a real dollar cost, so you always know what happened and what it cost, down to the call.

No black box, no surprise bill.

Cost and behavior are recorded on every run, not estimated after the fact.

Cost on every node

Every AI call is priced in real cents, per node and per run. You see what the orchestrator cost, what each worker cost, and what the whole run cost, not a single number at month end.

Priced from live model rates

Cost is computed from a model registry that syncs current provider pricing, and every run records which model actually ran. Switch models and the numbers follow.

Trends over time

A dashboard shows cost and usage over the last 7, 30, or 90 days, with the change against the prior period, so a creeping bill is obvious early.

Watch every run live

Nodes light up as they execute and stream their output, including the model's reasoning. You see what each agent did while it's doing it.

Replay any run

Every run is saved as a snapshot. Open it to see the exact graph, each node's input and output, status, timing, and cost. Filter the history by status.

Drill into a node

Open a node in a run to see its tokens, model, finish reason, cost, and timing. Multi-agent runs get an orchestrator view across all the workers.

Answers questions like

  • Which node is driving most of our spend?
  • Did this week cost more than last week?
  • What model did this run use, and why did it cost that much?
  • Where did this run slow down?
  • Which run failed last night, and at which node?

The bill matches the trace. The same per-run cost you watch in the dashboard is the cost you're billed on. Usage-based pricing means the number you see and the number you pay are the same number. See pricing.