The MindLab platform is a coordinated ecosystem of specialist agents. Each agent is a typed specialist, with a clear, machine-readable contract that specifies its capabilities, constraints, and operational parameters.

​
Agent Types

There are three main types of agents in the MindLab platform:
  • Planner Agents: These agents are responsible for decomposing high-level objectives into a directed acyclic graph (DAG) of smaller, well-defined sub-tasks.
  • Solver Agents: These agents are responsible for executing the core tasks in a workflow. They may call multiple tools and services to get the job done.
  • Reviewer Agents: These agents are responsible for checking the work of the solver agents against a predefined quality rubric. They can also be used to enforce consistency and to flag errors or hallucinations.

​
The Agent Lifecycle

Agents in the MindLab platform go through a lifecycle that is similar to the lifecycle of a human employee.
  1. Provisioning: Agents are provisioned from the Expert Registry.
  2. Configuration: Agents are configured to meet the specific needs of a team or a project.
  3. Execution: Agents execute tasks as part of a workflow.
  4. Evaluation: The performance of agents is continuously evaluated.
  5. Deprovisioning: Agents are deprovisioned when they are no longer needed.