README.md

metadata

title: DCI-Agent Demo
emoji: 🔮
colorFrom: purple
colorTo: indigo
sdk: docker
app_port: 7860
pinned: false

DCI-Agent Search

A Prestige Purple-themed Hugging Face Spaces app for autonomous document search over Wikipedia and other datasets.

Features

• Autonomous search: Pi reads files and runs shell commands (rg, grep, find, etc.) to answer your questions over a corpus.
• Real-time terminal: Watch Pi's reasoning and tool execution in a live terminal-like log.
• Default corpus: Wikipedia corpus with multi-hop QA examples from 2WikiMultiHopQA, HotpotQA, and other datasets.
• Upload your own corpus: ZIP upload with per-file (1 KB) and total (15 MB) limits.
• Native PI runner: Uses the original DCI-Agent Pi CLI via RPC. The Docker build now requires pi-mono to build successfully so remote deployments stay aligned with local behavior.

Deployment

Hugging Face Spaces (Docker)

1. Create a new Space on Hugging Face, choose Docker as the SDK.
2. Upload the contents of this web-app/ directory.
3. Set your OPENAI_API_KEY as a Space secret (optional — users can also provide it in the UI).
4. The Space will build and launch automatically.

Local testing

cd web-app
pip install -r requirements.txt
python app.py

Open http://localhost:7860 in your browser.

Usage

1. Enter your OpenAI API key.
2. Choose a model (gpt-4o recommended).
3. Set max turns (default: 6).
4. Either check "Use Default Corpus" or upload your own ZIP file.
5. Type a question and click Run Search.
6. Watch the terminal log for real-time progress.
7. The final answer appears in the Final Answer box.

File structure

web-app/
├── app.py                  # Gradio UI and orchestration
├── pi_wrapper.py           # PI RPC client and native runner integration
├── Dockerfile              # HF Spaces Docker image
├── requirements.txt        # Python dependencies
├── default_corpus/         # Wikipedia corpus (wiki_dump.jsonl)
│   ├── biology/
│   ├── earth_science/
│   ├── economics/
│   └── robotics/
└── README.md

Design

Prestige Purple theme:

• Deep purples (#B4AFF2, #5822B4) and rich blacks (#0A0A12, #141420).
• Plus Jakarta Sans for headings and body.
• JetBrains Mono for the terminal log.
• Cinematic gradients and generous whitespace.