Dispatch Studio-built components from a runner-only Agent

StudioRunnerTools is the dispatch half of the Studio: list the components in the platform database and run one by id.

StudioRunnerTools is the dispatch half of the Studio: list the components in the platform database and run one by id. It carries no create/edit/archive surface, so a router or team lead can hand work to built components without holding the Studio’s mutation tools. Runs execute as the current user, keep one session per component per conversation, and relay PAUSED results.

studio_runner_dispatcher.py

"""
Dispatch Studio-built components from a runner-only Agent
=========================================================

StudioRunnerTools is the dispatch half of the Studio: list the components in
the platform database and run one by id. It carries no create/edit/archive
surface, so a router or team lead can hand work to built components without
holding the Studio's mutation tools. Runs execute as the current user, keep
one session per component per conversation, and relay PAUSED results.

Prerequisites: OPENAI_API_KEY
Run: .venvs/demo/bin/python cookbook/05_agent_os/22_studio/studio_runner_dispatcher.py
Try: ask the dispatcher to run the same component twice and compare session ids
"""

import json
from pathlib import Path

from agno.agent import Agent
from agno.db.sqlite import SqliteDb
from agno.models.openai import OpenAIResponses
from agno.registry import Registry
from agno.tools.studio import StudioTools
from agno.tools.studio_runner import StudioRunnerTools

# ---------------------------------------------------------------------------
# Create a component with StudioTools
# ---------------------------------------------------------------------------

DB_DIR = Path(__file__).parent / "tmp"
DB_DIR.mkdir(exist_ok=True)
DB_FILE = DB_DIR / "studio_runner.db"
DB_FILE.unlink(missing_ok=True)

db = SqliteDb(
    id="studio-runner-db",
    db_file=str(DB_FILE),
)

registry = Registry(
    name="Runner Registry",
    models=[OpenAIResponses(id="gpt-5.5")],
    dbs=[db],
)

builder = StudioTools(registry=registry, db=db, default_model_id="gpt-5.5")
# Creates are drafts by default and dispatch only resolves the published
# version, so publish=True makes the component runnable immediately. Every
# StudioTools call returns a StudioResult envelope: read data on success.
created = json.loads(
    builder.create_agent(
        name="Haiku Writer",
        instructions="Answer with a single haiku.",
        model_id="gpt-5.5",
        publish=True,
    )
)
if not created["ok"]:
    raise RuntimeError(f"create_agent failed: {created['error']['code']}")

# ---------------------------------------------------------------------------
# Run it from a runner-only Agent
# ---------------------------------------------------------------------------

dispatcher = Agent(
    name="Dispatcher",
    model=OpenAIResponses(id="gpt-5.5"),
    tools=[StudioRunnerTools(registry=registry, db=db)],
    instructions="Discover what exists, then delegate the request to the right component.",
    db=db,
    markdown=True,
)


def main() -> None:
    dispatcher.print_response("Have the haiku writer produce a haiku about databases.")


if __name__ == "__main__":
    main()

Run the Example

1. Set up your virtual environment

   Mac

   uv venv --python 3.12
   source .venv/bin/activate
   
2. Install dependencies

   uv pip install -U "agno[scheduler]" openai sqlalchemy
   
3. Export your OpenAI API key

   Mac/Linux

   export OPENAI_API_KEY="your_openai_api_key_here"
   
4. Run the example

   Save the code above as studio_runner_dispatcher.py, then run:

   python studio_runner_dispatcher.py
   

Full source: cookbook/05_agent_os/22_studio/studio_runner_dispatcher.py