Agentsnew

Agent Memory

Manage how agents reuse context and local state between runs.

Overview

Agents in Overseer are focused workers. Each one combines instructions, a model choice, and a bounded toolset so the orchestrator has something predictable to delegate to.

Operator Note

Overseer works best when each layer stays explicit: the orchestrator routes, supervisors execute, tools return evidence, and traces show what actually happened.

Getting Started

Open the Agents Workshop, inspect the seeded agents, then narrow each one to the smallest useful scope before adding more capability.

Configuration

Keep prompts concrete, keep tool lists short, and prefer role clarity over cleverness.

Examples

agent.json

{
  "name": "release-analyst",
  "model": "qwen2.5:7b-instruct",
  "responsibility": "Summarize release activity and surface operator-facing issues",
  "tools": ["charting"]
}

Next Steps

Continue with Supervisor Concepts or head back to the documentation home.