Your team's compute,
under control.

No. Creating a company workspace and inviting members is free. Each key gets its own plan from $19/month when you're ready to pay for its compute, so a team can mix Starter, Economy, Standard, Pro and Pro Plus. Keys you can't size yet can run on pay as you go.

Every member gets their own API key, and each key has its own plan and monthly compute budget. One person's agent loop can't drain anyone else's budget.

That key's requests stop until the plan renews or an admin moves it to a bigger plan. Every other key keeps working. Plans have no overage charges.

When an admin approves the first plan. A member picks a plan for their key, and an admin approves it and pays through Stripe Checkout. Upgrades are prorated; downgrades apply at renewal. The workspace and invitations are free.

Yes. Compliance restricts inference to Europe, the United States, or both, for the whole company, a team or a single key. A team or key can narrow the company rule but never widen it. Restricted keys only reach approved provider endpoints with zero data retention and no training, and a request that can't be served there fails instead of going elsewhere. It controls where providers run inference; it is not an end-to-end data residency guarantee.

Yes. Each member uses their own key with any agent that supports our OpenAI-compatible endpoint. Setup guides cover the supported tools; have each member verify their real workflow before you size up.

Your company workspace is its own Stripe customer. Each key with a plan is its own subscription with its own monthly invoice, showing your company name, address and VAT number. Pay-as-you-go credit is bought by the company, and each top-up is invoiced too. Personal plans stay separate from the company.

Each developer gets a key with its own plan, and Claude Opus 5.5 and Opus 5 are on every plan. Six developers on Standard cost 6 x $89 = $534 a month with $95 of compute each; six on Pro cost 6 x $249 = $1,494 with $269 each. The budget is drawn at each model's per-token rate (Opus 5.5 lists at $4 / $20 per million tokens), so a developer who keeps Opus pinned all day uses a Standard budget well before the month ends. Routing hard steps to Opus and routine ones to efficient models is what makes a per-seat plan cover Opus-class work, and a key that runs out stops with HTTP 402 instead of adding to the invoice.

With a plan per key the forecast is a multiplication: developers x plan price x 12. Ten developers on Economy ($39) come to $4,680 a year, and ten on Standard ($89) to $10,680. Each key's invoice is its plan price every month, because requests stop at the key's compute budget instead of billing overage; the only change is a key an admin moves to a bigger plan. Pay as you go, at usage plus 15% from a prepaid company balance, is there for keys you can't size yet and stays off unless you turn it on.

A key per developer instead of one shared key, so one person's afternoon can't stop the other three. Four keys on Standard cost 4 x $89 = $356 a month with $95 of compute each, or four on Economy $156 with $41 each. Cline takes each key through its OpenAI Compatible provider at https://api.stdcmpt.com/v1, Claude models are on every plan, and there is no 5-hour or weekly window. A developer who keeps Sonnet pinned all day can use $95 before the month ends; routing routine steps to efficient models stretches it, and a key at its budget stops with HTTP 402 rather than billing more. Setup is free, and billing starts when an admin approves the first plan.

Yes, by giving each developer or tool its own key. Every key has its own plan and budget, and an admin can open Usage and the request log for a single key, which shows the model that served each request and its latency. A key that reaches its budget stops with HTTP 402 without touching the others, so a runaway tool shows up as one key at 100% rather than a mystery total.

Three layers: the agents your developers already use (Cline, Kilo Code, Claude Code or Codex CLI), one gateway key per developer instead of one shared key, and a model policy that sends routine steps to efficient models and hard ones to a frontier model. On Standard Compute that is six keys on monthly plans: six on Standard cost 6 x $89 = $534 a month (USD) with $95 of compute each, and a developer who needs more sits on Pro at $249 with $269. Check OpenRouter's activity by model first: if most of the $900 went to Opus or GPT-5.6 Sol on routine turns, routing those turns to models such as GLM 5.3 or DeepSeek V4 Pro is where the saving is. Each key stops at its budget with HTTP 402 instead of growing the bill, and the request log shows the model that served every request.

Standard Compute meets four of the five, and the fifth is a no today: it has no SOC 2 report yet. The four: an OpenAI-compatible API at https://api.stdcmpt.com/v1 plus the Anthropic Messages API; per-seat pricing as one key per developer on its own monthly plan, for example 25 x $89 Standard = $2,225 a month (USD) with $95 of compute per developer; Claude Opus 5.5, Sonnet 5.5, GPT-5.6 and GPT-6 models on every plan; and monthly Stripe invoices in the company's name with address and VAT number. What procurement can check in writing now: Terms under Norwegian law with GDPR, a published sub-processor list, request logs that hold metadata rather than prompt or response text, opt-in Europe or United States inference rules and an audit log of compliance changes.

That's the Business tier: compute sized to your volume, an uptime SLA and custom contract and invoicing terms. Contact us and an engineer replies within a business day.