Bounty

A research system that turns online conversations into cited findings. Bounty collects posts, comments, and replies from YouTube, Reddit, TikTok, Instagram, and X; discovers rising topics via Google Trends; and uses LLM analysis to extract signals — pain points, adoption patterns, objections, belief shifts — each backed by quotes and source links. If evidence is thin, it says so.

Users: investors (unknown-unknown discovery, pain-point research around companies), marketers (creative angles, competitor mentions), product teams (feature gaps, user complaints). The engine is horizontal; investing is the first lens, not a hardwired filter.

Live at bountyapi.com/dashboard (token-gated).

Read this first

1. AGENTS.md — operating manual: architecture, the two-pipeline warning, commands, deploy flow, credentials map, hard rules, known-broken list. Any agent (or human) working in this repo must read this before making changes.
2. STATE.md — product philosophy, what's built, gaps, priority order.

Quickstart

python -m pytest tests/ -x -q          # 189+ tests, must be green before every push
python -m uvicorn app:app --port 8000  # local dev; BOUNTY_ENV=development bypasses token gate
# open http://localhost:8000/dashboard

Deploy: push to main → Railway auto-builds → bountyapi.com. Nothing else.

What this repo is NOT

• Not the x402/USDC data-API marketplace — that code exists but is deferred (see docs/legacy/)
• Not Singapore property/real-estate tooling — legacy, deferred
• Not an MCP directory play — legacy, deferred

Old strategy and marketing documents live in docs/legacy/ and describe that earlier direction. They are kept for history only.

Layout