Configuration

YAAAF is configured through a JSON file that defines the LLM client, available agents, data sources, and external tools.

Configuration File

Set the YAAAF_CONFIG environment variable to point to your configuration:

export YAAAF_CONFIG=/path/to/config.json
python -m yaaaf backend

Basic Structure

{
  "client": {
    "model": "qwen2.5:32b",
    "temperature": 0.7,
    "max_tokens": 1024,
    "host": "http://localhost:11434"
  },
  "agents": [
    "sql",
    "visualization",
    "answerer",
    "websearch"
  ],
  "sources": [],
  "tools": []
}

Client Configuration

The client section configures the default LLM settings:

{
  "client": {
    "model": "qwen2.5:32b",
    "temperature": 0.7,
    "max_tokens": 1024,
    "host": "http://localhost:11434",
    "disable_thinking": false
  }
}

Field	Description	Default
`model`	Ollama model name	qwen2.5:32b
`temperature`	Creativity (0.0-2.0)	0.7
`max_tokens`	Maximum response length	1024
`host`	Ollama server URL	http://localhost:11434
`disable_thinking`	Disable extended thinking	false

Agent Configuration

The agents array lists which agents are available. Agents can be specified as strings (using defaults) or objects (with custom settings).

Simple Format

{
  "agents": ["sql", "visualization", "answerer", "websearch"]
}

Per-Agent Settings

Override model settings for specific agents:

{
  "agents": [
    "sql",
    {
      "name": "visualization",
      "model": "qwen2.5-coder:32b",
      "temperature": 0.1
    },
    {
      "name": "answerer",
      "model": "qwen2.5:32b",
      "temperature": 0.7,
      "max_tokens": 4096
    },
    {
      "name": "document_retriever",
      "host": "http://remote-ollama:11434"
    }
  ]
}

Available agent names:

sql - Database queries
visualization - Charts and graphs
answerer - Synthesize answers
reviewer - Analyze artifacts
websearch - DuckDuckGo search
brave_search - Brave Search API
document_retriever - Document search
url - URL content extraction
url_reviewer - URL content synthesis
bash - Filesystem operations
tool - MCP tool integration
mle - Machine learning
numerical_sequences - Data structuring
user_input - Interactive input

Data Sources

The sources array configures data sources for agents.

SQLite Sources

For SqlAgent:

{
  "sources": [
    {
      "name": "sales_database",
      "type": "sqlite",
      "path": "./data/sales.db"
    }
  ]
}

Text Sources

For DocumentRetrieverAgent:

{
  "sources": [
    {
      "name": "documentation",
      "type": "text",
      "path": "./docs/",
      "description": "Product documentation"
    }
  ]
}

Supported text formats: .txt, .md, .html, .htm, .pdf

RAG Sources

For persistent RAG storage:

{
  "sources": [
    {
      "name": "knowledge_base",
      "type": "rag",
      "path": "./data/rag_index.pkl",
      "description": "Persistent knowledge base"
    }
  ]
}

MCP Tools Configuration

The tools array configures external tools via Model Context Protocol (MCP).

SSE Tools

HTTP-based MCP servers:

{
  "tools": [
    {
      "name": "calculator",
      "type": "sse",
      "description": "Mathematical calculations",
      "url": "http://localhost:8080/sse"
    }
  ]
}

Stdio Tools

Command-line MCP servers:

{
  "tools": [
    {
      "name": "file_tools",
      "type": "stdio",
      "description": "File manipulation",
      "command": "python",
      "args": ["-m", "my_mcp_server"]
    }
  ]
}

API Keys

Some agents require API keys:

{
  "api_keys": {
    "brave_search_api_key": "YOUR_API_KEY"
  }
}

Get a Brave Search API key at https://api.search.brave.com/

Safety Filter

Configure content filtering:

{
  "safety_filter": {
    "enabled": true,
    "blocked_keywords": ["harmful", "dangerous"],
    "blocked_patterns": ["pattern.*regex"],
    "custom_message": "I cannot process that request."
  }
}

Complete Example

{
  "client": {
    "model": "qwen2.5:32b",
    "temperature": 0.7,
    "max_tokens": 2048,
    "host": "http://localhost:11434"
  },
  "agents": [
    "sql",
    {
      "name": "visualization",
      "model": "qwen2.5-coder:32b",
      "temperature": 0.1
    },
    "answerer",
    "reviewer",
    "websearch",
    "document_retriever",
    "tool"
  ],
  "sources": [
    {
      "name": "main_database",
      "type": "sqlite",
      "path": "./data/main.db"
    },
    {
      "name": "documentation",
      "type": "text",
      "path": "./docs/",
      "description": "Product documentation"
    }
  ],
  "tools": [
    {
      "name": "calculator",
      "type": "sse",
      "url": "http://localhost:8080/sse",
      "description": "Math operations"
    }
  ],
  "api_keys": {
    "brave_search_api_key": "YOUR_KEY_HERE"
  }
}

Configuration Generator

Use the interactive configuration generator:

python -m yaaaf config

This guides you through:

LLM model selection
Agent configuration
Source setup
Tool configuration

Environment Variables

Variable	Description
`YAAAF_CONFIG`	Path to configuration JSON file
`YAAAF_DEBUG`	Enable debug logging (true/false)
`BRAVE_SEARCH_API_KEY`	Brave Search API key (alternative to config)

Troubleshooting

Agent not available: Ensure the agent is listed in the agents array.

Source not found: Check that the path exists and is accessible.

Tool connection failed: Verify the MCP server is running and accessible.

Model not available: Pull the model with ollama pull model_name.