Hey Product Hunt, I built Kage because AI coding agents keep forgetting repo lore. Every new session starts with the same painful loop: where is this logic, why is this workaround here, which tests matter, what broke last time, and why did another agent already make this decision? Kage turns that knowledge into repo-local memory. The important part: Kage does not just save notes. It connects memory to code. Example: an agent fixes a flaky checkout retry test and discovers two retry paths look duplicated, but are intentionally different. One path retries external callbacks using idempotency keys. The other retries user checkout using session state. Kage saves that as a memory packet and links it to the retry modules and tests. Two weeks later, someone opens a fresh agent session and asks: "Clean up this duplicated retry logic." A normal agent may blindly refactor it. With Kage, the agent recalls: "This duplication is intentional. Here is why. These are the tests to run." That is the gap Kage focuses on. Agents do not just need more context. They need the right repo knowledge at the moment they touch the relevant code. Kage is open-source, local-first, git-visible, CLI + MCP ready, and designed for teammates to share repo memory through the repo itself. I would love feedback from people using coding agents seriously: would you commit this kind of repo memory with your code? Website: https://kage-core.com GitHub: https://github.com/kage-core/Kage npm: https://www.npmjs.com/package/@k...

Kage

Shared repo memory for AI coding agents

Shared repo memory for AI coding agents