Agent Flywheel
Capture agent work with runtime SDKs and plugins, then use Mission Control to generate, train, evaluate, and improve agents with approval controls.
Improve agents from real work
Capture agent activity in the runtime, send evidence to Mission Control, and approve evaluated improvements before release.
Agent runtime → SDK/plugin capture → outbound evidence → Mission Control: tasks → grading → training → evaluation → approval
Mission Control runs the task, grading, training, evaluation, and approval workflow server-side. Local SDKs and plugins capture and send evidence; they do not manage rollout approval.
1. Capture
Runtime SDKs and plugins collect agent activity and tool evidence.
2. Improve
Mission Control turns captured evidence into tasks, grades results, and trains agents.
3. Approve
Mission Control evaluates candidate changes and manages approval before promotion.
Responsibilities
| Component | Responsibility |
|---|---|
| Runtime | Runs the agent and produces activity and tool evidence. |
| Mission Control | Handles server-side tasks, grading, training, evaluation, and approval. |
| Agent Skill | Lets operators manage server-side training from an agent workflow. |
Explore
Quickstart
Capture your first agent work and follow the initial workflow.
Concepts
Understand how evidence moves through tasks, grading, training, and evaluation.
Integrations
Connect runtime SDKs and plugins to capture agent activity and tool evidence.
Authentication
Configure access for agents and workflow management.