Features · Costs and limits

See what every step costs,and never get billed past your plan.

Every run shows what each step cost and which AI model it used, and every workflow tells you what a run should cost before it starts. When your plan’s monthly credits run out, runs stop. You’re never billed past your plan.

Free plan, no card needed.

The Usage tab of Runs: execution credits left this month, then model spend, calls and tokens, and the cost by model.

The Usage page in a demo workspace: credits left this month, AI model spend, calls and tokens, and cost by model.

Step by step

Where the money goes, run by run.

  1. Step 1 of 6

    Know the cost before it runs

    Every workflow says what a run should cost before you start it: the credits, and the AI spend from your model wallet, from its cheapest path to its most expensive.
    • On Run now and in the workflow’s settings
  2. Step 2 of 6

    Every run shows its cost

    The Runs page lists every run with the steps it took, how long it ran, the tokens it used and what it cost, newest first.
    • The Usage page adds it all up by model and by day, with the credits left this month
  3. Step 3 of 6

    Every step shows its cost

    Open a run to see each step’s AI model and cost. Use a small, cheap model where a step only sorts or checks, and a strong one where judgment matters.
    • A step that does work costs 1 credit, whichever model it uses
    • Branches and other logic or data steps cost 0.1 credit
    • Waiting for a person’s approval costs nothing, however long it takes
  4. Step 4 of 6

    Many steps need no AI

    Branches and Python steps (Team plan and up on Cloud) make decisions without an AI model, so you only pay for AI where it’s actually needed.
  5. Step 5 of 6

    It stops at your cap

    Give a workflow a cap per run, in credits or in AI dollars on Cloud. The run stops before the step that would go over it and says so, with a button to raise the cap.
    • A step that’s already working finishes first
  6. Step 6 of 6

    Credits plus AI spend

    Credits pay for running workflows. AI models are paid separately, from your model wallet, or by your provider if you use your own keys. Every plan includes $2 a month of AI models from us.
    • Warnings when 50%, 25% and 10% of the month’s credits are left
    • Buy more credits on a paid plan, or move up a plan
    • With your own keys, we charge nothing for AI models
Run now opened on the Weekly supplier review workflow: before the run starts it says each run costs about 2.7 credits and $0.0004–$0.0005 from your model wallet.
  1. Step 1 of 6

    Run now opened on the Weekly supplier review workflow: before the run starts it says each run costs about 2.7 credits and $0.0004–$0.0005 from your model wallet.

    Know the cost before it runs

    Every workflow says what a run should cost before you start it: the credits, and the AI spend from your model wallet, from its cheapest path to its most expensive.
    • On Run now and in the workflow’s settings
  2. Step 2 of 6

    The Runs page, newest first and grouped by day: each run's workflow, what started it, the path of its steps, how long it took, its tokens, cost and start time, with chips to filter by status, a sort and search. Logs, Evals and Usage are tabs beside it.

    Every run shows its cost

    The Runs page lists every run with the steps it took, how long it ran, the tokens it used and what it cost, newest first.
    • The Usage page adds it all up by model and by day, with the credits left this month
  3. Step 3 of 6

    A completed Weekly supplier review run: its time, steps, tokens and cost, that it followed the saved workflow, and its path as a timeline: the parallel search and agent, the merge, and the review it wrote.

    Every step shows its cost

    Open a run to see each step’s AI model and cost. Use a small, cheap model where a step only sorts or checks, and a strong one where judgment matters.
    • A step that does work costs 1 credit, whichever model it uses
    • Branches and other logic or data steps cost 0.1 credit
    • Waiting for a person’s approval costs nothing, however long it takes
  4. Step 4 of 6

    Invoice triage on the canvas: Check totals (Transform) feeds Needs review? (If/Else), which goes to Accounts payable approval (Human) when true and straight to Draft payment note (LLM) when false. Add step and Tidy up sit in the dock below.

    Many steps need no AI

    Branches and Python steps (Team plan and up on Cloud) make decisions without an AI model, so you only pay for AI where it’s actually needed.
  5. Step 5 of 6

    An Order lookup run stopped by its cap: “This run reached its cap of 1 credit”, with Raise the cap and Replay. Its first step finished; the two after it were not reached.

    It stops at your cap

    Give a workflow a cap per run, in credits or in AI dollars on Cloud. The run stops before the step that would go over it and says so, with a button to raise the cap.
    • A step that’s already working finishes first
  6. Step 6 of 6

    Settings, Billing: execution credits and the model wallet, automatic top-up, and credit history.

    Credits plus AI spend

    Credits pay for running workflows. AI models are paid separately, from your model wallet, or by your provider if you use your own keys. Every plan includes $2 a month of AI models from us.
    • Warnings when 50%, 25% and 10% of the month’s credits are left
    • Buy more credits on a paid plan, or move up a plan
    • With your own keys, we charge nothing for AI models

What it’s for

What builders do with these numbers.

Price a job before you quote it
The estimate shows what a run will cost, from its cheapest path to its most expensive, before it ever runs.
Find the step that costs the most
Every run shows each step’s model and cost, so you know which step to move to a cheaper model.
Cap what a client’s workflow can spend
Set a cap per run, in credits or in AI dollars on Cloud. The run stops before the step that would go over it.
Use your own API keys
Bring your own AI provider key, and your provider bills you for the models, with nothing extra from us.

A worked example

A weekly competitor check: $1.36 a month in AI costs, or $0.20.

The job: every Monday, read five competitors’ pricing and changelog pages, and write the team a note when something changed. Priced at list prices: Claude Sonnet 4.6 at $3 per million tokens read and $15 per million written, Claude Haiku 4.5 at $1 and $5.

One agent does it all: $1.36 a month
It reads all five pages in full every Monday and writes the note: about 103,500 tokens read and 2,000 written on Sonnet. $0.34 a run.
A workflow does it: $0.20 a month
Plain steps fetch the pages and keep only the parts that matter, with no AI model. Haiku checks whether anything changed, and Sonnet writes only when it did. $0.09 a run with a change, $0.016 without.
About 6.7 times cheaper
Same job, same note. Even in a month where every Monday has a change, the workflow costs $0.34, still four times less.
Every sum, shown
The token counts, the prices and each step’s share are in the post why a workflow costs less than asking an AI agent every time. On Cloud, the Python step it uses comes with the Team plan and up.

Your rules

What it never does to your bill.

It never bills past your plan
When the month’s credits are used up, runs stop until the next month. You’re never billed beyond the plan you chose.
It checks between steps
A run that’s already going may finish a small shortfall, up to 5% of the month’s credits. A bigger one stops before the step it can’t pay for, and says why.
It tells you the cost first
Each workflow shows what a run should cost, from its cheapest path to its most expensive, in its settings and on Run now.
It stops at your cap
Set a cap per run, in credits or AI dollars on Cloud. The run stops before the step that would go over it; a step already working finishes.
Waiting is free
A run waiting for a person, or queued for a free slot, costs nothing while it waits.
Runaway runs stop
Each plan has a longest run time. It’s a safety guard, not an allowance: a run past it ends and says why.

Plans

What a credit is, on every plan.

A step
An agent, an AI model call, an evaluator, a tool or a nested workflow costs 1 credit. A logic or data step costs 0.1. A person’s step costs nothing.
A run
Its steps added up, or one credit per started minute of active work if that’s more, and never less than one credit.
On your own servers
No credits at all: your AI providers bill you directly, and the estimate and a cap in AI dollars work the same way.

See pricingThe full guide in the docs

Know the bill before it arrives.

Let us run it, with nothing to install and no API key needed. Or run it on your own servers, free and open source.

Free plan, no card needed.