Agent crews, runlike infrastructure.
Describe the team once — researcher, analyst, writer — and Knacto runs it on demand. Every model call lands in a ledger you can open, question, and bill against.
Visual editor; the code path is optional
One click, a schedule, or POST /run
Unused reserve refunded
Mix OpenAI, Anthropic and Gemini in one crew
Give each agent a role, a goal, and the tools it may touch.
Order them in a pipeline, or set a manager to delegate.
Fire from the dashboard, on a schedule, or from your own code.
Read the per-step receipt; unspent reserve comes back.
Sixty seconds, one crew, real work.
Build the crew, wire the flow, press run, read the receipt — the whole loop in one take.
Six crews you can clone right now.
Each one ships wired — agents, tasks, process. Swap the models and press run.
Market brief
Scans a topic, weighs drivers against risks, returns a one-page brief with sources.
Longform pipeline
Keyword in, edited SEO article out — outline, draft, edit, fact-check.
Triage desk
A manager agent reads the issue, orders a repro, and posts the fix plan.
Account dossier
Company name in — firmographics, a fit score, and a first-touch email out.
Filing digest
Reads a quarterly filing, pulls the numbers that moved, flags the surprises.
Support drafter
Sorts the ticket, finds the right doc, drafts a reply for a human to send.
Market brief
A trace you can bill against.
Open any run and see who did what: the model, the tool calls, the token draw per step. Costs are reserved before a run and settled after — the ledger never surprises you.
One crew, one endpoint.
Anything you build in the editor is callable from your stack the moment you save it. Same queue, same metering, same receipt.
- →Scoped keys per project and environment
- →A webhook fires on finish, carrying the full receipt
- →Tools metered per call: search, fetch, code runner
Top up once. No clock running.
Packs are one-time purchases. Bigger packs carry bonus tokens.
Sign-up comes with 5,000 tokens on the house. Runs settle against a reserve — finish under it and the difference returns to your balance.
Before you ask.
What exactly draws tokens?
Model calls and tool calls, per step. The rate is flat; a typical three-to-four-agent run lands between 400 and 700 tokens, and the trace shows the draw line by line.
What happens if a run stops early?
You're only charged for the steps that actually executed. The rest of the reserve goes straight back to your balance — no ticket, no waiting.
Can two agents in one crew use different models?
Yes — that's the point. A cheap, fast model for gathering, a stronger one for the final write-up. OpenAI, Anthropic and Gemini can sit side by side; bring your own keys or run on ours.
How are scheduled and API runs billed?
Identically to dashboard runs: same reserve, same metering, same receipt. The trigger changes; the ledger doesn't.
Is there anything to install?
No. Crews are built and run in the browser; the queue does the heavy lifting. The API and the sandboxed code runner are there when you want them — not a requirement.