AI-native product discovery that connects your code, docs, data, and user research — finds where they disagree — and turns conflicts into agent-ready specifications.
Cursor and Claude Code accelerated how teams build software. But someone still has to figure out what to build. That process is still manual, fragmented, and disconnected from the coding agents that do the building.
Every product has four sources of truth that are constantly in tension. PMs add value where they conflict.
What CAN happen?
What's EXPECTED?
What IS happening?
What SHOULD happen?
Compass ingests your product's sources, finds where they conflict, and generates actionable specifications.
Point Compass at your code, docs, analytics, interviews, and support tickets. It extracts structured evidence from each.
compass connect && compass ingestCompass cross-references all sources to find conflicts, then ranks evidence-grounded product opportunities.
compass reconcile && compass discoverGenerate agent-ready specifications with task breakdowns, file paths, and acceptance criteria — formatted for Cursor or Claude Code.
compass specify "opportunity title"Every recommendation cites specific evidence. No generic advice. No hallucinated priorities.
See exactly where your sources of truth disagree — strategy says mobile-first but code investment says desktop, users ask for features the roadmap doesn't mention.
Product opportunities ranked by evidence strength. HIGH confidence means 3+ sources corroborate. Every recommendation traces back to specific evidence.
Feature specifications formatted for Cursor and Claude Code — with task breakdowns, file paths, acceptance criteria, and testing requirements.
Desktop app with built-in AI. Also available as a CLI and MCP server.
$ pip install compass-ai