How are cloud runs charged?

By running time, from your plan’s included usage first and from the wallet only once the pool is spent: $0.002 per minute on shared cloud compute, $0.003 per minute on the isolated tier. A run that fails is still metered for the time it ran.

One flat rate for every model: $3.20 per million tokens, with no per-call fee. It draws from the same pool as compute — the per-plan token figure in usage reporting is telemetry, not a free band. If you bring your own AI keys on your own agent, no managed AI is billed.

Do local runs cost compute?

No. A run on your own local agent is not gated, not metered and not charged — free and unlimited on every plan, with parallelism bounded by your machine, not the plan.

Does my wallet balance expire?

No. The wallet holds only credit you purchase, and it never expires. The included credit pool is separate: it is the plan price in credits, it resets monthly, and it is always spent before the wallet is touched.

docs ▸Operate ▸Billing and usage

Runs onWrit Cloud

billing ▸ one pool

Billing & usage

Every plan includes one monthly pool of included usage — the plan’s price expressed in credits, at 1 credit = $0.001. Cloud compute, residential egress and AI all draw from that one pool, pool first; the wallet holds only credit you purchased, and is touched only once the pool is spent. Runs on your own local agent are free, on every plan.

pool ▸ the plan price, in credits

One pool, three meters

There are no separate allowances to forecast — no “cloud minutes” bucket next to an “AI tokens” bucket. The month’s pool funds whatever mix the month actually needs, and resets each billing cycle:

Plan	Monthly pool	In short
Free	1,000 credits · $1	Try it out on your own machine.
Starter	15,000 credits · $15	For individuals shipping their first integrations.
Pro	49,000 credits · $49	For client-grade APIs on authenticated sites.
Growth	199,000 credits · $199	For teams running automations in production.
Scale	499,000 credits · $499	For high-volume, multi-team operations.
Enterprise	800,000 credits · $800	Contact-only: custom terms on the same one-pool model.

Spending is grant-first: as long as the pool has credit, usage consumes the pool and the wallet is untouched.

rates ▸ what draws from it

The rates

Resource	Rate	Notes
Cloud compute — shared	$0.002/min (2 credits/min)	Billed for the time the run actually used, success or failure.
Cloud compute — isolated	$0.003/min (3 credits/min)	1.5× shared: the run gets a private, single-use sandbox.
Residential egress	$7/GB	Pay-as-you-go, by the gigabyte actually moved.
AI	$3.20 per 1M tokens	One flat rate for every model, with no per-call fee.
Local / self-hosted agent	Free on every plan — no compute charge, no metering (details below).

AI is not a free band. The per-plan AI token figure you may see in usage reporting is telemetry — it describes what you used, not an allowance. AI meters from the pool like everything else, at the flat rate above. With your own provider keys on your own agent, no managed AI is billed at all.

wallet ▸ overage only

The wallet

The wallet holds only credit you purchased — the included pool never passes through it. When the pool is spent, metered usage continues from the wallet at exactly the same rates: adding funds buys more usage, never a different price. Add between $5 and $1000 at a time (presets $10 · $25 · $50 · $100). Purchased credit never expires.

Worked example

A cloud run that takes 30 seconds, and the same workflow on your own machine:

cloud    30 s = 0.5 min × $0.002/min  = $0.001
         + AI, per token, at $3.20/1M
         + residential, per GB, only if the run used it
─────────────────────────────────────────
local    same workflow, your own agent   = $0.00

empty ▸ what a 402 looks like

When pool and wallet are both empty

Cloud work stops instead of billing you. The request answers a structured 402 whose body carries a machine-readable code, your current usage, the limit it hit and an upgrade_hint — so your software can branch on it. The message offers the three ways out:

add funds to the wallet,
run it on your own local agent — unlimited and free,
or upgrade the plan.

Two notification events track the edge before you hit it: billing.low_balance and billing.credits_exhausted — the latter says it plainly: “Local agent runs are unaffected.”

local ▸ free by contract

Local is free — structurally

A run with execution_target=local is not gated, not metered and not charged. This is not a discount: billing checks simply do not apply to local runs, on any plan, including Free. Parallelism is bounded by your machine, not by the plan. Only cloud runs — and auto runs that land on cloud — are metered.

What is plan-limited is how many local agents you can keep linked to the platform at once:

Plan	Linked local agents
Free	1
Starter	1
Pro	1
Growth	5
Scale	20
Enterprise	Unlimited

caps ▸ per plan

Cloud duration caps

Cloud runs and live streaming sessions carry per-plan duration ceilings. Local runs have neither — the limits below are cloud limits:

Plan	Max cloud run	Max streaming session
Free	2 min	5 min
Starter	4 min	10 min
Pro	5 min	15 min
Growth	10 min	30 min
Scale	15 min	60 min
Enterprise	15 min	60 min

See pricing for the full plan matrix.

Usage, spend by category and invoices live on the billing page inside the app. Billing is an account surface, not part of the developer API — an API key carries authorization only, never the ability to read or move money.

What next

Where it runs: cloud, local and self-hosted, side by side.
AI sessions: what the token meter is attached to.
Pricing: plans and limits in one table.