Python API

The public API is intentionally small.

`create_deep_agent`

Use it when you want direct control over the agent:

from agno_deep_agents import create_deep_agent


agent = create_deep_agent(
    model="ollama:gemma4:e4b",
    instructions="Be practical and answer in English.",
)

agent.print_response("Analyze this project.", stream=True)

Common parameters:

Parameter	Description
`model`	Provider string or an `agno.models.base.Model` instance.
`tools`	Extra Python functions or Agno toolkits.
`instructions`	Extra instruction added to the team leader.
`workspace`	Root directory for file and shell tools.
`db_file`	SQLite path.
`skills_dir`	Local skills directory.
`ollama_host`	Custom Ollama host.
`enable_shell`	Enables or disables `run_shell`.
`allowed_shell_commands`	Command-name allow-list for shell tools.
`allow_all_shell_commands`	Disable command-name filtering while keeping workspace checks.
`compress_tool_results`	Enable Agno context compression for tool results.
`compression_model`	Optional model used only for compression.
`compression_tool_results_limit`	Compress after N uncompressed tool results.
`compression_token_limit`	Compress after a token threshold.
`send_media_to_model`	Send attached media to the model.
`store_media`	Store attached media in Agno session storage.
`max_iterations`	Team task-loop limit.
`show_members_responses`	Shows or hides specialist member responses.

`DeepAgentConfig`

Use it when configuration grows:

from agno_deep_agents import DeepAgentConfig, create_deep_agent


config = DeepAgentConfig(
    workspace=".",
    model="openai-responses:gpt-5.2",
    enable_shell=True,
    compress_tool_results=True,
    max_iterations=8,
)

agent = create_deep_agent(config)

`run_deep_agent`

Use it for direct execution:

from agno_deep_agents import run_deep_agent


run_deep_agent(
    "Map the project and recommend the next useful improvement.",
    user_id="dev@example.com",
    session_id="demo",
)

run_deep_agent also accepts multimodal input through DeepAgentMedia or direct images, audio, videos, and files keyword arguments:

from agno_deep_agents import DeepAgentConfig, DeepAgentMedia, run_deep_agent


run_deep_agent(
    "Describe this screenshot and suggest one UI fix.",
    DeepAgentConfig(model="openai-responses:gpt-5.2"),
    media=DeepAgentMedia(images=["screenshot.png"]),
)

Use a dedicated compression model when the main model is expensive:

from agno_deep_agents import DeepAgentConfig, create_deep_agent


agent = create_deep_agent(
    DeepAgentConfig(
        model="openai-responses:gpt-5.2",
        compression_model="openai-responses:gpt-4o-mini",
        compression_tool_results_limit=2,
    )
)

Custom Tools

Python functions can be passed through tools, following the Agno pattern:

from agno_deep_agents import create_deep_agent


def get_weather(city: str) -> str:
    """Get weather for a city."""
    return f"It's always sunny in {city}!"


agent = create_deep_agent(
    model="ollama:gemma4:e4b",
    tools=[get_weather],
)