How to use Fred

Current operating model#

Fred no longer assumes a built-in “leader agent” coordinating a crew.

The default pattern is:

• one selected agent per conversation
• one strong model profile for that agent step
• explicit tool calls for retrieval, actions, or external systems

This gives simpler behavior, clearer observability, and easier production hardening.

Which agent should you choose?#

• Choose a general ReAct agent for broad conversations with optional tools.
• Choose a domain agent (for example tracking, tabular, or reporting) when you need specialized prompts and tools.
• Prefer tool design over agent-to-agent delegation for most use cases.

Where configuration lives#

Fred uses YAML catalogs as source of truth:

• agentic-backend/config/configuration.yaml: platform/runtime settings (security, storage, endpoints)
• agentic-backend/config/agents_catalog.yaml: agent list, enablement, and class or definition_ref
• agentic-backend/config/models_catalog.yaml: model profiles, defaults, and routing rules

Minimal v2 catalog example#

# agents_catalog.yaml
version: v1
agents:
  - id: "ReActV2"
    name: "ReActV2"
    type: "agent"
    definition_ref: "v2.react.basic"
    enabled: true

# models_catalog.yaml
version: v1
default_profile_by_capability:
  chat: default.chat.openai.prod
profiles:
  - profile_id: default.chat.openai.prod
    capability: chat
    model:
      provider: openai
      name: gpt-5.1
rules: []

Run and verify#

1. Configure secrets in .env (OPENAI_API_KEY or your provider credentials).
2. Start the stack with make run.
3. Open the UI, start a conversation, and select a v2 agent.
4. Check backend logs to confirm selected model profile and tool calls.

Use the main repository quick start for the latest commands: https://github.com/ThalesGroup/fred#readme.

Extend with your own agent#

For a practical step-by-step guide using current v2 authoring (ReActAgent, @tool, YAML registration), see Add your agent.

The backend implementation lives in agentic-backend.