Mission Control documentation¶
Mission Control gives your applications governed access to AI models and MCP tools. Use your existing clients, control access by Team and virtual key, and inspect requests in one Console.
Make your first request Explore the concepts
New here? Choose your starting point for application development, managed Agents, Chat, administration or operations.
Start with what you want to build¶
Call AI models¶
Point an OpenAI client at your gateway, choose an available virtual model, and send a request. Stream responses, generate structured output, or let a model use governed tools.
Connect tools¶
Use your Team's permitted MCP tools from Claude Code or your own application. Mission Control checks access, applies rules, and asks for approval when a tool requires it.
Build and run agents¶
Run your existing agent through the gateways, or build and publish an agent in the Console. Use an ordinary service-account virtual key for external code, or publish a managed deployment with separate execution access.
Understand a request¶
Find a model or tool call by its returned Trace ID. Add application spans to see your code and gateway calls together, then group Agent conversation turns into Sessions.
Find the right guide¶
| You want to | Go to |
|---|---|
| Understand the system before building | Concepts |
| Complete a development or Console task | User guide |
| Measure Agent quality and compare versions | Evaluation workflow |
| Set up customer access and governance | Administration |
| Install and operate the services | Self-hosting |
Choose your client¶
| You use | Start here |
|---|---|
| OpenAI Python or TypeScript | OpenAI SDK integration |
| Claude Code | Connect with browser sign-in or a key |
| An MCP TypeScript application | MCP client integration |
| PydanticAI | Run an agent with models, tools, and traces |
| An application that emits OpenTelemetry | Send application traces |
The integration guides include setup, examples and recovery. The API reference describes supported endpoints and compatibility. Read How Mission Control works for the full picture or use the glossary for a term.
Set up and operate Mission Control¶
An administrator prepares Teams, models, tools, key policy and budgets before developers connect. An operator runs the services on your infrastructure.
- Set up an organization: invite people, configure providers and MCP servers, and enable resources.
- Self-host Mission Control: choose source, release-artifact or container deployment, then configure storage, networking, and backups.
- Diagnose a failed request or troubleshoot the services.
Release and design history¶
These guides describe the current release. The separate Product direction archive preserves planned capabilities and architecture decisions; its pages are excluded from documentation search and the current-guide AI index. Internal specifications and implementation plans are not published.