AgentOS showcase

This lesson assembles the curriculum into one coherent AgentOS process. It serves a pgvector-backed Agno documentation assistant, a current web-and-finance researcher, and a finance Team while storing sessions, traces, knowledge, and evaluation results in PostgreSQL.

The app always enables Bearer authentication with OS_SECURITY_KEY for protected REST requests. It also enables AgentOS tracing and runs one real AccuracyEval before the server starts.

Files

File	What it teaches
`_agents.py`	Shared PostgreSQL storage, pgvector documentation knowledge, Agno Assist, and Sage.
`_teams.py`	A finance Team combining Sage with a focused market-data specialist.
`demo.py`	One secured and traced AgentOS plus knowledge preparation and a stored accuracy evaluation.

Prerequisites

Install the demo environment with ./scripts/demo_setup.sh.
Start PostgreSQL with pgvector on port 5532:
bash
```
./cookbook/scripts/run_pgvector.sh
```
Export OPENAI_API_KEY for Agno Assist, embeddings, the finance specialist, and the Team leader.
Export ANTHROPIC_API_KEY for Sage and the accuracy judge.
Export a private OS_SECURITY_KEY; protected REST requests must send it as a Bearer token.
Keep internet access available for the Agno documentation page, web search, and Yahoo Finance.

Use a development key only for this local example:

bash

export OS_SECURITY_KEY="replace-with-a-local-secret"

Run

From the repository root:

bash

.venvs/demo/bin/python -m cookbook.05_agent_os.24_showcase.demo

Set SHOWCASE_PORT when port 7777 is already in use; otherwise the app uses port 7777.

Startup performs three observable steps:

Insert https://docs.agno.com/agent-os/introduction.md into the PostgreSQL content table and pgvector table, using skip_if_exists=True to avoid a duplicate row when the URL-derived content ID is already stored.
Query pgvector and require at least one documentation match.
Run AgentOS Documentation Accuracy with Agno Assist and a Claude judge, then store the result before serving port 7777 by default.

The distinct Uvicorn module target is cookbook.05_agent_os.24_showcase.demo:app.

Authenticate and discover the app

An anonymous request is rejected:

bash

curl -i http://127.0.0.1:7777/config

The root page, GET /health, and GET /info remain public for discovery and health checks.

Supply the security key to discover the two Agents and one Team:

bash

curl \
  -H "Authorization: Bearer ${OS_SECURITY_KEY}" \
  http://127.0.0.1:7777/config

The same header is required for runs, sessions, knowledge, traces, and evaluation endpoints.

Prompts to try

Component	Prompt
Agno Assist	`Which three component types can AgentOS serve? Link the relevant documentation.`
Sage	`Give me current context and market data for NVDA, separating facts from interpretation.`
Finance Team	`Compare MSFT and GOOGL fundamentals and recent news in a concise table.`

Agno Assist is the RAG path: it searches the indexed documentation before answering. Sage uses WebSearchTools for current context and YFinanceTools for market data. The Finance Team delegates to Sage and its market specialist, then reconciles both member responses.

Inspect the result

After connecting http://127.0.0.1:7777 to the AgentOS control plane with the security key, inspect:

Agents and Teams for agno-assist, sage, and finance-team.
Knowledge for the Agno Documentation instance and indexed introduction.
Sessions after trying the three prompts.
Traces for the Agent, model, tool, and Team spans created by those runs.
Evals for the stored AgentOS Documentation Accuracy result.

The equivalent authenticated REST surfaces are GET /config, GET /knowledge/content, GET /sessions, GET /traces, and GET /eval-runs.

Shutdown

Stop the server with Ctrl-C. The PostgreSQL data remains available for the next run; stop the shared pgvector container only if you started it solely for this lesson.