---
title: DAO — Data Agent Observatory
emoji: 🔬
colorFrom: blue
colorTo: purple
sdk: docker
app_port: 8501
pinned: false
tags:
  - streamlit
  - docker
  - data-agents
  - observability
  - kdd
---

# DAO — Data Agent Observatory

DAO — Data Agent Observatory is team1647's creative-track submission for making data-agent execution inspectable, reproducible, and judge-friendly.

This Hugging Face Space runs the DAO Streamlit application from the packaged team1647 submission. DAO turns benchmark run artifacts into an observability cockpit where reviewers can inspect run quality, task reasoning, evidence, replay, confidence, reliability, and future-proofing workflows without treating the data agent as a black box.

## Demo Mode

This Space is designed to support a demo-first review flow.

If precomputed run artifacts are present, reviewers can inspect:

1. **Run Intelligence** — run-level quality, reliability, resource usage, failures, validation, and artifact completeness.
2. **Task Intelligence** — task-level reasoning, replay, evidence, confidence, verification, and raw trace inspection.
3. **Future Proof** — evidence-backed steering, checkpoint review, rerun planning, and comparison workflows.

Artifact inspection does not require API credentials.

## Live Execution

Live task execution requires model API secrets to be configured by the Space owner.

Expected secrets / variables include:

```text
MODEL_API_URL
MODEL_API_KEY
MODEL_NAME
```

Optional OpenAI-compatible variables may also be used if supported by the active config:

OPENAI_API_KEY
OPENAI_API_BASE
OPENAI_MODEL

If secrets are not configured, the Space should still allow inspection of bundled demo artifacts.

Streamlit Entry Point

The application starts from:

src/data_agent_baseline/observatory/app.py

The container serves Streamlit on:

0.0.0.0:8501

Hugging Face provides HTTPS externally, so the container does not need to generate or serve a self-signed certificate.

Suggested Review Flow
Open the Space.
Select a bundled run if available.
Start with Run Intelligence for aggregate run-level diagnosis.
Open Task Intelligence for task-level reasoning, replay, and evidence inspection.
Open Future Proof for evidence packs, steering, rerun planning, and comparisons.
Use live execution only if model secrets are configured.