Pick the models. Set the caps. Manage AI usage and spend. Your team keeps the tools they already use.
Allow models for the whole organization, one team, or one person. New models stay blocked until you approve them.
*-preview
—
Blocked at the top means blocked everywhere.
rolling 7 days, not a calendar week
org.hosted_budget_request.approve · actor admin · policy v42
Approve all of it, or just some.
Weekly caps per person, per team, and org-wide. Hosted requests stop at the cap. People ask for more; you approve in one click.
Who ran what, on which model, and what it cost. One view.
Same week, by provider, model, or team.
Dashboard, CSV, or a scoped API token. Your FinOps stack plugs straight in.
Continuum speaks the OpenAI and Anthropic APIs, streaming included. Codex, Claude Code, and the official SDKs just work.
Every member gets their own key, so caps and usage attach to a person.
No SDK swap, no proxy config, no code change. One variable.
Anthropic
verified
Sealed on arrival. Never shown back.
Bring your Anthropic and OpenAI keys as a pass-through, or use ours. Same controls either way.
Every tool call, every file, every URL. On the record.
Org-wide allow and block lists come with Enterprise.
Logged today. Block lists with Enterprise.
Four teams, one weekly dollar axis, $3,260 total. Each tick is that lead's ceiling: $2,000, $1,500, $1,000, $600.
Invite people, appoint leads, set their limits. Leads approve within them.
Offboarding cuts access, keys, and devices in one step.
Direct traffic runs laptop to provider. We see usage, not content.
Remote control is end-to-end encrypted. Everything stored is encrypted and scoped to your organization. Security
Your key, our route. Policy and budget are checked before the request leaves, which is what makes the cap hard.
Your keys, your subscriptions, laptop to provider. We see usage, and we act inside your Anthropic and OpenAI accounts. Real control, and a smaller kind of it.
$25 / member / month
Hosted. The base weekly allowance.
5 seats × $25 = $125 / month
Get Team Everything in the personal plans, plus:Let's talk
Book a demo Everything in Team, plus:Short answers here. The long ones are in the docs.
Yes. Policy runs organization, then team, then member, with allowlists and deny patterns at each scope. A lower scope can narrow access but never re-open what a higher scope denied, and new models stay blocked until an admin reviews them.
No. One subscription, billed by your live member count, with an included weekly allowance per tier: $25 a week on Plus, up to $1,000 a week on Ultra. Past the allowance, overage is prepaid at cost with no markup, and a member with overage switched off simply stops at the cap.
No, not today. Sign-in and invitations run on WorkOS AuthKit, with Google and Apple login, and there is no screen for federating your own provider or syncing a directory. Removing someone in WorkOS is what triggers offboarding: provider access blocked, devices and tokens revoked, content key destroyed.
Yes on hosted, no on direct. Hosted proxies content through our gateway on purpose, because that is what makes a cap enforceable. Direct and BYOK sessions run machine to provider and send us usage only. Transcript sync, if you turn it on, stores pages encrypted at rest under a per-member key. Encrypted, not zero knowledge.
No. Hosted cap decisions are admin-only by design. A lead can approve a subscription request from their own team, up to their ceiling, and set budgets within the ceiling an admin gave them; past that the backend refuses.
Yes. An admin adds your Anthropic and OpenAI admin keys; each is verified before storage, sealed under its own data key, and never shown again beyond a fingerprint. Provisioning and enforcement act through those keys, so the contract stays yours.
No, not as a self-serve setting today. Every session captures the tool calls an agent made and the URLs it fetched; org-wide allow and block lists are scoped as part of an Enterprise engagement.
Yes, as a scoped Enterprise engagement rather than a download. One offer scopes a private deployment of the same single-tenant stack we run in production. The other adds the hardware: we spec, purchase, and deploy a GPU cluster on NVIDIA, AMD Instinct, or Cerebras, then commission your gateway on it.
One subscription, billed by your live member count. Set the models, the caps, and the roles once your people are in.
model policy · weekly caps · approvals · compatible apis · enterprise engagements