Researching a product or topic means sifting through hundreds of Reddit threads, Hacker News discussions, and web articles — most of it noise. Research Assistant is an AI agent that does this for you. It collects real user opinions from across the web, scores every comment for relevancy, and synthesizes the results into a cited summary.
Research Assistant runs a three-stage agent pipeline:
- Collection Agent — Searches Reddit, Hacker News, Product Hunt, and the web for threads and articles matching your query. Extracts comments and quotes from each source.
- Scoring Agent — Reads every collected comment and scores it 1–10 for relevancy to your research question using an LLM. Assigns categories in product mode. Filters out noise so only substantive content reaches the summary.
- Summary Agent — Reads through the top-scored comments and generates a cited summary, weighting findings by both AI relevancy scores and community upvotes. Every claim links back to its source.
You stay in control throughout — star comments, override AI scores, adjust summary focus, and regenerate individual sections with feedback.
- General Research — Enter any question or topic. The agent pipeline searches across sources, scores comments, and generates a single cited summary.
- Product Research — Enter a product name. The agents run targeted searches across six categories (issues, feature requests, competitors, benefits, alternatives, general info) and generate a structured report with per-category summaries.
pip3 install -r requirements.txt- Reddit: Create a "script" app at https://www.reddit.com/prefs/apps — note the client ID and secret
- OpenAI: Get an API key at https://platform.openai.com/api-keys
- Product Hunt (optional): Get a Developer Token at https://www.producthunt.com/v2/oauth/applications
cp .env.example .envEdit .env with your credentials:
REDDIT_CLIENT_ID=your_client_id
REDDIT_CLIENT_SECRET=your_client_secret
OPENAI_API_KEY=sk-your-key-here
PRODUCT_HUNT_API_TOKEN=your_developer_token_here # optional
python3 app.pyOpen http://localhost:5000.
- Choose General Research or Product Research on the homepage
- Enter your question or product name, adjust settings (sources, max threads/comments, time range), and click Research
- Watch the live activity feed as threads and comments are collected and scored
- Browse the sortable Threads and Comments tables — click a thread to filter its comments and view the full post
- Star interesting comments, set your own relevancy scores, and filter by source
- Click Summarize (or Generate Summaries in product mode) for AI-generated summaries with numbered citations
- Use customize to control comment count and provide focus instructions
- In product mode, regenerate individual summary cards with per-card feedback
- Click Find More Comments & Articles to expand your results, or Export CSV to download
| Variable | Default | Description |
|---|---|---|
REDDIT_CLIENT_ID |
(required) | Reddit app client ID |
REDDIT_CLIENT_SECRET |
(required) | Reddit app client secret |
REDDIT_USER_AGENT |
ResearchAssistant/1.0 |
User agent for Reddit API |
OPENAI_API_KEY |
(required) | OpenAI API key |
PRODUCT_HUNT_API_TOKEN |
(optional) | Product Hunt Developer Token |
LLM_MODEL |
gpt-4o-mini |
OpenAI model for the scoring agent |
ALT_SUMMARY_MODEL |
gpt-4.1-mini |
Alternate model for the summary agent (togglable in UI) |
PORT |
5000 |
Port to run the app on |
Uses GPT-4o-mini for scoring and summarization by default. Typical cost: ~$0.02-0.05 per general research query, ~$0.10-0.15 per product research (more searches and 6 summary calls). You can use a stronger model like gpt-4.1 for summaries via ALT_SUMMARY_MODEL.
Research data is stored in data/research.db (SQLite). CSV exports are saved to data/exports/. The data/ directory is created automatically and is git-ignored.