Configuration

Customize models, agent behavior, API keys, and deployment settings.

Environment Variables

Agent Zero reads configuration from a .env file mounted at /a0/.env inside the container. Variables prefixed with A0_SET_ override Agent Zero's internal settings.

API Keys

Variable	Description	Required
`API_KEY_OPENROUTER`	OpenRouter API key for LLM access	Yes
`OPENAI_API_KEY`	OpenAI API key (for embeddings)	Recommended

Model Settings

Variable	Default	Description
`A0_SET_chat_model_provider`	`openrouter`	Provider for chat/reasoning model
`A0_SET_chat_model_name`	`moonshotai/kimi-k2.5`	Chat model name
`A0_SET_util_model_provider`	`openrouter`	Provider for utility/fast model
`A0_SET_util_model_name`	`google/gemini-2.5-flash-preview`	Utility model name
`A0_SET_embed_model_provider`	`openai`	Provider for embeddings
`A0_SET_embed_model_name`	`text-embedding-3-small`	Embedding model name
`A0_SET_chat_model_ctx_length`	`128000`	Max context length for chat model

Auth & Security

Variable	Default	Description
`AUTH_LOGIN`	`admin`	Web UI login username
`AUTH_PASSWORD`	—	Web UI login password
`A0_SET_mcp_server_token`	(auto-generated)	API key for `X-API-KEY` auth

Note: By default, mcp_server_token is auto-generated at startup. The included a0-dotenv/patch-settings.py script patches Agent Zero to respect the token set in .env, enabling fixed API keys for backend integration.

Feature Toggles

Variable	Default	Description
`A0_SET_a2a_server_enabled`	`true`	Enable Agent-to-Agent communication
`A0_SET_mcp_server_enabled`	`false`	Enable MCP server
`A0_SET_update_check_enabled`	`false`	Disable auto-update checks

Agent Profiles

Each agent's behavior is configured in agents/bmad-{role}/:

settings.json

Controls model parameters specific to each agent:

{
    "chat_model_kwargs": {"temperature": 0.5},
    "chat_model_ctx_length": 128000,
    "chat_model_ctx_history": 0.75
}

Field	Description
`temperature`	Controls randomness. Low (0.2) = precise; High (0.7) = creative
`chat_model_ctx_length`	Maximum tokens in the context window
`chat_model_ctx_history`	Fraction of context reserved for conversation history (0.0–1.0)

prompts/

Contains the system prompt that defines the agent's persona, behavior, and instructions. Edit these to customize how each agent responds.

SKILL.md

Defines the agent's skill for Agent Zero's skill system — triggers, capabilities, inputs, and outputs. This enables the Orchestrator to route tasks to the right agent.

Switching Models

To use a different model, update the a0-dotenv/.env file:

# Example: Use Claude Sonnet via OpenRouter
A0_SET_chat_model_provider=openrouter
A0_SET_chat_model_name=anthropic/claude-sonnet-4-20250514

# Example: Use GPT-4o via OpenRouter
A0_SET_chat_model_provider=openrouter
A0_SET_chat_model_name=openai/gpt-4o

# Example: Use Qwen 3.5 via OpenRouter
A0_SET_chat_model_provider=openrouter
A0_SET_chat_model_name=qwen/qwen3-235b-a22b

Then restart the containers:

docker compose -f docker-compose.live.yml restart

GMeet Backend Config

The GMeet backend reads from environment variables or defaults:

Variable	Default	Description
`GMEET_PORT`	`4000`	REST API port
`GMEET_WS_PORT`	`4001`	WebSocket port
`A0_API_KEY`	`bmad-live-api-2026`	API key for Agent Zero authentication
`DATABASE_URL`	PostgreSQL connection string	Optional persistence backend