Control model spend and access for every team
People and agents get a virtual credential with their team's budget and approved models. Switch providers in one place.
See where your model spend goes
Break down spend by person, team, and model across closed and open models. Set a budget for each team.
Claude Code
CodexClaude Desktop
AnthropicOpenAI
FireworksVirtual keys for every team
Add your OpenAI, Anthropic, or open-model key once. It backs every team's virtual key and never leaves the gateway.
A team's key only reaches the models you approve. Expensive or unapproved models stay off the list.
Each key carries its team's monthly cap. Calls stop at the cap and every budget change is logged.
Agent Monitor shows each call's provider, model, cost, and latency next to the agent's tool calls.
Cap spend and switch models anytime
A budget for each team
Each team's budget follows its SSO group and stays current as people join and leave.
Track costs without storing prompts
With zero data retention on, the gateway logs model, tokens, cost, and latency for each call. Prompts and responses are not stored.
Change providers in one place
Move a team to a new provider or model with one policy change.
Virtual credentials for every user
People and agents get a virtual credential to the gateway endpoint. Your provider keys stay in the gateway. Revoke one person's access without rotating anything.
Know what every team spends on models
Add a provider key, create a policy for one team, and point a laptop at the gateway.
Frequently asked questions
Claude Code, Codex, Claude Desktop, other common agent harnesses, and coworker agents. Test locally with the included installer or roll it out to your team through MDM.
OpenAI, Anthropic, open models (e.g., Kimi, GLM, DeepSeek), and any OpenAI-compatible or Anthropic-compatible endpoint. Connect each with an API key or link your team's existing subscription seats.
Yes. Connect it as an OpenAI- or Anthropic-compatible endpoint and the gateway adds team budgets and model controls on top.
Calls stop at the team's cap. Admins can raise it for the team or lift it for individual users.
Yes. Add your negotiated rates and the gateway prices each call with them.
Yes. Grant a policy to a directory group and access updates as people join or leave it. You can also grant individual users access.
The MCP gateway controls which tools an agent can use. The LLM gateway controls which models it can call and how much it spends. Both share the same users, groups, and audit log.