How to use Fred

Who is Fred?#

Fred is an open-source, agentic flow chatbot designed to coordinate a team of expert agents. It supports complex user queries by dynamically planning and routing tasks to specialized sub-agents.

You can:

• Ask Fred directly (he’ll act as a team coordinator).
• Or interact with a single expert for focused, domain-specific questions.

When to Use Fred vs. Experts#

How Fred is Configured#

The agentic part of Fred is configured via a YAML file (typically config/configuration.yaml). This file lists the available agents, their roles, and their model/tool configurations.

A simplified example:

ai:
  default_model:
    provider: openai
    name: gpt-4o

  agents:
    - name: Fred
      class_path: app.agents.leader.leader.Leader
      type: leader
      enabled: true

    - name: GeneralistExpert
      class_path: app.agents.generalist.generalist_expert.GeneralistExpert
      enabled: true

    - name: TabularExpert
      class_path: app.agents.tabular.tabular_expert.TabularExpert
      enabled: true
      mcp_servers:
        - name: knowledge-flow-mcp-server
          url: http://localhost:8111/mcp_tabular
          transport: sse

    - name: DocumentsExpert
      class_path: app.agents.documents.documents_expert.DocumentsExpert
      enabled: true
      mcp_servers:
        - name: knowledge-flow-mcp-server
          url: http://localhost:8111/mcp_text
          transport: sse

👉 Tip: Refer to the live configuration file in fred/backend/configuration.yaml.

Trying Fred#

1. Enable experts in the YAML config.
2. Set your API key (OPENAI_API_KEY=... in .env).
3. Run backend + frontend using make run.
4. Optionally, run Knowledge Flow for RAG/document support.
5. Start the UI

See the Getting Started guide for a live walkthrough.

Experiment & Extend#

• Try asking Fred complex questions to observe multi-agent collaboration.
• Ask experts like TabularExpert or DocumentsExpert directly for focused queries.
• Watch logs to understand reasoning steps, expert selection, and tool usage.

Want to Create Your Own Expert?#

• Create a class extending AgentFlow.
• Implement async_init() with:
  • Your model (get_model(...))
  • Tool loading if needed
  • Prompt + graph definition
• Register the agent in your YAML config (or via API).
• ✅ Done!

For details about the agent architecture, see the agentic_backend README.

🤝 Contribute#

Fred is an open, extensible platform. Feel free to open issues, suggest improvements, or build your own agents and share them with the community!