Skip to content

User guide

Choose the task you want to complete. These guides use an existing deployment; Self-hosting covers installation and service operations.

New to Mission Control? Complete a getting-started tutorial first. For explanations of access, Agents and evaluation, use Concepts.

Models and tools

Agents and Chat

Task Guide
Create and manage a managed application Create and manage an Agent
Authenticate code as a user or service account Issue and manage virtual keys
Keep your agent in your own process Run an external agent
Draft, test, and publish an agent in the Console Build a definition
Have a conversation with a published agent Chat with an Agent
Invoke a published agent from your application Embed an Agent

Observe and diagnose

Evaluate quality

Follow the evaluation workflow from a first dataset to production quality checks.

Task Guide
Define what a good answer means Define and record scores
Collect repeatable test cases Create and curate a dataset
Measure an Agent version Run an experiment
Automate grading Create a grader
Compare models or changes Compare evaluation evidence
Monitor production quality Sample production traffic
Review judgments and enforce a release threshold Human review and publish gates
Run evaluation from automation Evaluation in CI

Sign in and get access

Follow Organizations and sign-in, Teams, models, and virtual keys, and Request access. Each task states its prerequisites; see Roles and permissions for the eight built-in presets and bounded custom roles and their exact permissions.

The workspace header keeps Organization and Team context visible. Overview is a direct destination for every Organization member; capabilities determine Access, Observe, Build and Manage destinations. Teams and Virtual keys own setup and first-request handoff. See context and navigation for scoped links, Organization-wide page indicators, context switching, collapsed navigation and the mobile drawer. Unknown/unauthorized resources use a non-disclosing unavailable state.