​
Payment method

There is no free tier. Every project, sandbox included, needs a card on file and a plan before it can do billable work. In the console, open Platform API → Billing, pick the project and choose a plan: Checkout runs on Stripe. When it completes, the project’s card is on file within a few seconds (the page waits for Stripe’s confirmation). Manage billing opens Stripe’s customer portal to change the card, see invoices or cancel. Choosing the other plan later replaces the subscription: the old one ends at once and its usage so far is invoiced. Until a card is on file, billable requests answer 402 and do nothing:
That covers creating agents, POST /v1/conversations/{id}/messages, calls and realtime sessions (/v1/calls), inbound calls to your numbers, POST /v1/messages, POST /v1/numbers, channel connections, /v1/computers (create, start and toolset calls), and /v1/chat/completions with a project key. Reading data, accounts, model keys, webhooks and the spend cap keep working, so you can set up before you add a card. The same 402 payment_method_required comes back when a project’s subscription is cancelled, or when a renewal payment fails and Stripe stops retrying (unpaid). While Stripe is still retrying (past_due) work continues; update the card in Manage billing. Growth credit. Each paid Growth subscription invoice (the first one and every renewal) adds a $300 Stripe credit grant to the project’s customer, applied to that period’s metered usage. It doesn’t carry over. Each project has its own Stripe customer, so one project’s usage never lands on another’s bill.

​
Spend cap

Every project has a monthly spend cap: the most it can spend in one calendar month (UTC). New projects start at $100. When the project’s spend reaches the cap, billable requests are refused until you raise the cap or the month ends.

​
How spend is counted

Spend is this month’s usage at the prices on that page: each meter’s quantity times its price, plus the token rows at the model’s list price + 15%. It is counted at list price, before any plan credit or discount; those are applied on your invoice. Spend is updated as usage is recorded, so it is current to the last request.

​
What the cap stops

At or over the cap, the billable requests listed under Payment method answer 402 with {"error": {"type": "billing_error", "code": "spend_cap_reached"}} and do nothing (sandbox texts and simulated numbers cost nothing and aren’t refused by the cap). Reading data, managing agents, accounts and webhooks, and receiving texts keep working. One reply or text that starts under the cap can take the project a little past it; the next one is refused. If the spend check itself fails, the request is refused with 503 spend_check_unavailable rather than run unchecked. Retry it.

​
Read or lower the cap

Scope usage.
Scope usage, from a key that isn’t bound to an account. An API key can only lower the cap. Raising it answers 403 spend_cap_raise_in_console: raise it in the console at platform.allternit.com (project settings), so a leaked key can’t lift your limit. spend_cap_cents is 0 to 100,000,000; 0 stops all billable work.

​
Threshold webhooks

Subscribe a webhook to usage.threshold to hear when spend reaches 50%, 80% and 100% of the cap. Each fires at most once per project per month:
used and limit are in US dollars. Lowering the cap below what you already spent sends the thresholds you just crossed.

​
Plans

Request and concurrency limits are enforced per project (see Idempotency and limits).