Showcase Demo

Load Agno documentation into pgvector, run and store one real AccuracyEval, then serve two Agents and one finance Team with authentication and tracing.

demo.py

"""
Run the Complete AgentOS Showcase
=================================

Load Agno documentation into pgvector, run and store one real AccuracyEval,
then serve two Agents and one finance Team with authentication and tracing.

Prerequisites: Postgres on port 5532, OPENAI_API_KEY, ANTHROPIC_API_KEY, and OS_SECURITY_KEY
Run: .venvs/demo/bin/python -m cookbook.05_agent_os.24_showcase.demo
Try: authenticate to GET /config, then run agno-assist, sage, or finance-team
"""

import os
from importlib import import_module

from agno.eval.accuracy import AccuracyEval, AccuracyResult
from agno.models.anthropic import Claude
from agno.os import AgentOS
from agno.os.settings import AgnoAPISettings

agents_module = import_module("cookbook.05_agent_os.24_showcase._agents")
teams_module = import_module("cookbook.05_agent_os.24_showcase._teams")

AGNO_DOCS_URL = agents_module.AGNO_DOCS_URL
agno_assist = agents_module.agno_assist
agno_docs = agents_module.agno_docs
sage = agents_module.sage
showcase_db = agents_module.showcase_db
finance_team = teams_module.finance_team

security_key = os.getenv("OS_SECURITY_KEY")
if not security_key:
    raise ValueError("OS_SECURITY_KEY is required for the authenticated showcase")
showcase_port = int(os.getenv("SHOWCASE_PORT", "7777"))

# ---------------------------------------------------------------------------
# Create Secured and Traced AgentOS
# ---------------------------------------------------------------------------

agent_os = AgentOS(
    id="agent-os-showcase",
    name="AgentOS Showcase",
    description="A secured, traced AgentOS with RAG, research, finance, and evaluation.",
    db=showcase_db,
    agents=[agno_assist, sage],
    teams=[finance_team],
    knowledge=[agno_docs],
    tracing=True,
    settings=AgnoAPISettings(os_security_key=security_key),
)
app = agent_os.get_app()

accuracy_evaluation = AccuracyEval(
    db=showcase_db,
    name="AgentOS Documentation Accuracy",
    model=Claude(id="claude-sonnet-4-6"),
    agent=agno_assist,
    input="Which three component types can AgentOS serve?",
    expected_output="AgentOS can serve Agents, Teams, and Workflows.",
    additional_guidelines=[
        "Accept capitalization and singular or plural variations.",
        "The answer must identify all three component types.",
    ],
    num_iterations=1,
    show_spinner=False,
)


def prepare_knowledge() -> int:
    """Load the docs page and prove pgvector can retrieve it."""
    agno_docs.insert(
        name="Agno introduction",
        url=AGNO_DOCS_URL,
        skip_if_exists=True,
    )
    matches = agno_docs.search(
        "Which components can AgentOS serve?",
        max_results=3,
    )
    if not matches:
        raise RuntimeError("The Agno documentation load produced no vector matches")
    print(f"Knowledge search returned {len(matches)} pgvector match(es)")
    return len(matches)


def run_accuracy_evaluation() -> AccuracyResult:
    """Run the live judge evaluation and require one stored result."""
    result = accuracy_evaluation.run(
        print_summary=True,
        print_results=False,
    )
    if result is None or not result.results:
        raise RuntimeError("The showcase AccuracyEval produced no result")
    if showcase_db.get_eval_run(result.run_id) is None:
        raise RuntimeError("The showcase AccuracyEval was not stored")
    print(
        f"AccuracyEval stored {len(result.results)} result(s); "
        f"average score={result.avg_score}"
    )
    return result


# ---------------------------------------------------------------------------
# Run the Showcase
# ---------------------------------------------------------------------------

if __name__ == "__main__":
    prepare_knowledge()
    run_accuracy_evaluation()
    agent_os.serve(
        app="cookbook.05_agent_os.24_showcase.demo:app",
        host="127.0.0.1",
        port=showcase_port,
    )

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[os]" "psycopg[binary]" anthropic ddgs openai pgvector yfinance
   
3. Export environment variables

   Mac/Linux

   export ANTHROPIC_API_KEY="your_anthropic_api_key_here"
   export OPENAI_API_KEY="your_openai_api_key_here"
   export OS_SECURITY_KEY="your_os_security_key_here"
   
4. Run PgVector

   macOS / Linux

   docker run -d \
     -e POSTGRES_DB=ai \
     -e POSTGRES_USER=ai \
     -e POSTGRES_PASSWORD=ai \
     -e PGDATA=/var/lib/postgresql \
     -v pgvolume:/var/lib/postgresql \
     -p 5532:5432 \
     --name pgvector \
     agnohq/pgvector:18
   
5. Clone Agno

   Clone the pinned Agno source and run the remaining commands from its root:

   git clone https://github.com/agno-agi/agno.git
   cd agno
   git checkout v3.0.4
   
6. Run the example

   Run the example from the repository root:

   python -m cookbook.05_agent_os.24_showcase.demo
   

Full source: cookbook/05_agent_os/24_showcase/demo.py