v1.0.23: Atlas and hardening

Adds the Atlas knowledge graph with Carts (semantic map, cards, timeline, saved views, and an MCP cart tool), compact float32 embeddings, and a launch-hardening security pass across the API, auth, and MCP surfaces. Ships the /replen skill and @replen/mcp, now in the official MCP Registry.

v1.0.13: Self-tuning matching

Adds the self-tuning learning loop (relevance floor, adaptive lookback, triage suppression), k-anonymous cross-user quality signals, and the Watchtower watch lenses (dependency releases and security advisories in your stack). Matching sharpens the more you triage.

v1.0.0: First milestone

The foundational Replen release: the discovery pipeline, source fetchers, and dashboard. Replen is a local intelligence layer for AI coding agents. It maps each of your repos to its technical capabilities and watches the open-source ecosystem for what fits yours. The heavy reasoning runs in your own AI session on your tokens, so there is no per-user API bill and your source never leaves your machine. Try it with npx replen.