Launching today
Lock
The decision protocol for product teams and AI agents
10 followers
The decision protocol for product teams and AI agents
10 followers
Your AI coding agent just shipped a fix that contradicts a decision your team made six months ago in Slack. Nobody noticed. Lock captures product decisions where they happen, in Slack and in the terminal, then flags when a new one contradicts an old one. The GitHub Action catches conflicts in PRs, including the ones your agent opens. Auto-classified, tracked through supersession, searchable by meaning. The decision layer for teams shipping with agents.
Hi Product Hunt,
I'm Philippe. I work in product. The reason I built Lock is specific to this moment.
Two years ago, "why did we build it this way?" was a question you could ask the engineer who shipped the feature. Today that engineer is half-pairing with Claude Code or Cursor, and half the decisions were made in a ten-minute agent session that nobody else saw. The PR lands. The transcript closes. The decision is gone.
Volume of decisions: up. Visibility of decisions: down.
The problem I was trying to solve: capture product decisions wherever they're made, before they disappear. ADRs, Notion pages, Linear descriptions all share the same failure mode. The decision happens in one place. The documentation lives somewhere else. Nobody writes the second step. In the agentic era this gap is bigger, not smaller, because more decisions get made and fewer of them are made by people who own the documentation.
How Lock works: one storage, multiple capture surfaces.
- Slack bot: `@lock ` in any channel, or extract from a thread
- CLI: `lock ""` from the terminal, where humans review agent output
- GitHub Action: comments on PRs with relevant past decisions, including the PRs opened by agents
Under the hood: PostgreSQL with pgvector, auto-classification by type (product, technical, business, design, process), conflict detection against past decisions, semantic search, supersession chains. The conflict detection is the part that gets more important the faster your team ships. A contradicting decision should be caught at PR time, not three months later when someone asks why.
Where it's at: early. Functional, but small. MIT licensed, self-hosted, telemetry off by default. Two stars on GitHub today, which I'd rather you raise because you tried it than because you upvoted.
What I'd love feedback on:
1. For teams using Claude Code or Cursor: where do your agent-era product decisions live right now? PRs? Closed transcripts? Vibes?
2. What would have to be true about Lock for it to be the first place your team writes a decision down, not the last?
Site: https://www.uselock.ai/
Happy to answer anything.
Philippe