Download OpenWork
AI GatewayKeys stay on the server

Add a provider once.Everyone gets the right models.

Connect OpenAI, Anthropic, Bedrock, Vertex or Azure in OpenWork Cloud. Choose who can use each model and set spend limits. Members pick a model and never handle a key.

Set up AI GatewayRead the docs
On OpenWork Cloud or self-hosted.
AI Providers
  • Anthropic
  • OpenAI
  • Amazon Bedrock
  • Google Vertex AI
  • OpenRouter
Who can use modelsMembers can't add their own keys.
Spend limitsThis month
$3,240Spent by 48 people. Keys stay on the server.
  • Everyone$31 avg of $50 / person
  • Engineering
  • Design team
    Warning sent at 90%. Pauses at 100%.

A working preview. Add a provider, change who can use it, raise a limit.

01

Add a provider

Anthropicsk-ant-•••••••••••• Shared API keyAdd provider
02

Choose who can use it

Who can use itEngineeringModelsClaude Sonnet, HaikuMembers' own keysNot allowed
03

Members pick a model

From Acme StudioClaude Sonnet Claude Haiku
How a prompt reaches a model

One key for each person.Provider keys never leave the server.

Priya in OpenWork
modelClaude Sonnetkeyow_gw_••••••••> Summarize the Q3 pipeline
AI Gateway
Key belongs to PriyaEngineering can use Claude Sonnet$148 of $200 used this month
  • Anthropic● Sent with the organization key
  • OpenAIOrganization key
  • Google Vertex AIPriya's Google sign-in

Keys stay on the server

Provider keys are stored encrypted and only the gateway reads them. The app gets one AI Gateway key for each person.

Access checked on every request

Grant models to everyone, a team or one person. Remove someone and their next request is refused.

Usage by person

Tokens and estimated cost for the last 31 days, by model, team and person.

Spend limits

Set the budget before the month starts.

Every request goes through the gateway, so a limit applies in whichever app a member uses.

EngineeringAnthropic, OpenAI
Daily
$20
Monthly
$200
Used this month
$148 of $200
When someone reaches it
Pause their models
Let them ask for 25% more
On

Daily, weekly and monthly

Give a team or a person an amount in US dollars for each period. A member has to stay under all of them.

Pause or only warn

Stop requests when someone reaches a limit, or let them keep working and flag it for follow-up.

Members can ask for more

Let members ask for 25% more when they reach a limit. An admin approves or denies the request.

Providers

Use the providers you already pay for.

Already run LiteLLM or Bifrost? Keep it and add it as a provider. Compare them

  • OpenAI
  • Anthropic
  • Google Vertex AI
  • Amazon Bedrock
  • Azure OpenAI
  • Mistral
  • OpenRouter
  • LiteLLM and more
Rolling out

More ways to use the same models.

Smart Model RouterRolling out on Cloud

Members pick Smart instead of a model. Each prompt goes to the model you tagged for that kind of work, at the thinking level it needs. How routing works

Keys for Claude Code, Codex and CursorRolling out

Members create their own AI Gateway API key and use your organization's models in other tools. The same access rules and limits apply. How API keys work

Pricing

Team$10 per user a monthShared provider keys and model access rules.
EnterpriseTalk to usSpend observability, audit log and SCIM.
OpenWork Models$10 per user a monthOptional. Managed models with no keys to add.

Your providers bill model usage. The same terms apply on OpenWork Cloud and self-hosted.

Frequently asked questions

Which providers does AI Gateway support?
OpenAI, Anthropic, Google Gemini, Google Vertex AI (including Claude on Vertex), Azure OpenAI, Microsoft Foundry (Claude), Amazon Bedrock, Mistral, OpenRouter, LiteLLM, and compatible catalog providers. The provider support page in the docs lists the credentials and APIs each one supports.
Do members ever see a provider key?
No. Provider keys are stored encrypted on the server and only the gateway reads them. Each member gets one AI Gateway key, which proves who is asking and only reaches the models that person is allowed to use.
How is this different from OpenWork Models or bringing your own keys?
AI Gateway routes your own providers through OpenWork with access rules, limits, and usage tracking. OpenWork Models is a managed subscription with no keys, billed separately. With your own keys in the desktop app, the app calls the provider directly and nothing is shared with a team.
How accurate is the cost view?
Costs are estimates from the tokens each provider reports and public model prices, or the cost a provider reports when it returns one. They are not your provider's bill.
What does AI Gateway cost?
Shared provider keys are on the Team plan at $10 per user per month. Spend observability and the audit log are on Enterprise. The same terms apply on OpenWork Cloud and self-hosted. Your providers bill model usage.
Can I keep LiteLLM or Bifrost?
Yes. Add your LiteLLM proxy as a provider and OpenWork syncs its models and teams. Other OpenAI-compatible gateways can be added as a custom provider.

Give your team models without handing out keys.

Set up AI GatewayRead the docs
Shared provider keys are on the Team plan.