Dossier is a web-based, multi-agent research desk. Enter a topic and it gathers web sources, extracts a source for more detail, drafts a report, and asks a reviewer model to critique the result. The interface displays each stage and lets you download the report.
The deployed service also exposes its health check at /health and interactive API documentation at /docs.
The frontend calls the backend in four stages:
- Search: Tavily returns up to five web results for the topic.
- Read: Tavily extracts the text from the leading result.
- Write: Groq generates a report from the search results and extracted text.
- Review: Groq critiques the report and suggests improvements.
The backend also provides POST /api/research, which runs a separate LangChain agent workflow using the search and scraping tools in tools.py, followed by the report writer and critic in agents.py.
- Backend: Python, FastAPI, Uvicorn, LangChain
- Frontend: HTML, CSS, and vanilla JavaScript
- Web search and extraction: Tavily Search and Extract APIs
- Text generation and review: Groq API via
langchain-groq; the current code selectsopenai/gpt-oss-120b - Browser assets: Google Fonts and jsPDF are loaded from external CDNs
The backend needs TAVILY_API_KEY and GROQ_API_KEY. Keep these keys on the server; do not put them in frontend code or commit your .env file.
Use Python 3.10 or later. From the project root, create a virtual environment and install the backend dependencies:
py -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install -r backend\requirements.txt
Copy-Item .env.example .envEdit .env and set your keys:
TAVILY_API_KEY=your_tavily_api_key
GROQ_API_KEY=your_groq_api_keyStart the server from the backend directory so its relative frontend paths resolve correctly:
Set-Location backend
python main.pyOpen http://127.0.0.1:8000. API documentation is available at http://127.0.0.1:8000/docs.
On macOS or Linux, the equivalent setup is:
python3 -m venv .venv
source .venv/bin/activate
python -m pip install -r backend/requirements.txt
cp .env.example .env
# Add TAVILY_API_KEY and GROQ_API_KEY to .env
cd backend
python main.pyAll endpoints are served by the same backend origin. JSON endpoints accept and return JSON.
| Method and path | Purpose | Request body |
|---|---|---|
GET /health |
Check whether the backend is responding. | None |
POST /api/search |
Search the web with Tavily (up to five results). | {"query":"research topic"} |
POST /api/extract |
Extract a web page with Tavily. | {"url":"https://example.com/article"} |
POST /api/chat |
Generate text with Groq. | {"messages":[{"role":"user","content":"Write a summary"}]} |
POST /api/research |
Run the complete agent-based research pipeline. | {"topic":"research topic"} |
/api/search returns a results array containing Tavily result data. /api/extract returns Tavily's extraction response, including extracted raw_content when available. /api/chat returns {"response":"..."}. /api/research returns search_results, scraped_content, report, and feedback.
The UI uses /api/search, /api/extract, and /api/chat for its staged progress display. The complete /api/research endpoint is available for direct API clients.
.
├── agents.py # Search/reader agents, report writer, and critic
├── pipeline.py # Command-line research pipeline
├── tools.py # Tavily search and web-page scraping tools
├── requirements.txt # Broader project/Python dependencies
├── .env.example # Environment variable template
├── backend/
│ ├── main.py # FastAPI app, API routes, and frontend serving
│ ├── requirements.txt # Backend runtime dependencies
│ ├── README.md # Backend-specific notes
│ └── test_resp.json # Example/test response data
└── frontend/
├── index.html # Research desk interface
├── style.css # Interface styles
├── script.js # UI behavior and backend requests
└── README.md # Frontend-specific notes
- Search, extraction, and model generation require valid API keys and may be subject to provider quotas or rate limits.
- The frontend requires an internet connection to load its Google Fonts and jsPDF CDN assets.
- Research reports are AI-generated summaries of retrieved material. Check the cited sources before relying on claims.