Xylem: Managing Agent Memory and Collaboration via Git
Rofl_im_jonny · reddit · 2026-07-13
This Reddit post introduces Xylem: a zero-dependency MCP stack for AI agent memory, collaboration, and knowledge retention.
The author's problems when multiple agents share a repository included:
- Context loss between sessions
- Multi-agent modifications to the same file easily overwriting each other
To solve this, they built three MCP servers:
- context-keeper: Saves project memory, decisions, and constraints, allowing new sessions to get up to speed quickly
- agentsync: Instead of relying on a database, it writes agent declarations to JSON files in dedicated branches, using git push as a lock; a failed push means someone else modified the reference first. The author ran 1,000 concurrent race condition tests with zero lost or duplicate grants
- cambium: Distills completed work into reusable experience, settling knowledge layer by layer from the individual to the team and then the organization
Its current retrieval is lexical: based on token overlap and tag weighting, without embeddings. The author mentions running benchmarks on query rewriting, which proved sufficient at the current scale, and notes they will switch to semantic retrieval if data proves it necessary.
The project can run locally or remotely via Cloudflare Workers, and it is already listed on the official MCP registry, making installation very low-friction.
More from coding & agent
- Travel MCP Server adds flight, hotel, weather and budget tools for agents — modelcontextprotocol · 2026-07-21
- OpenAI hackathon project stalls as Codex struggles on voice, while Claude spots the issue — ColleenMBrady · 2026-07-21
- Douyin Video Analysis MCP turns share links into structured video summaries — modelcontextprotocol · 2026-07-21
- An indie builder open-sourced 50+ AI apps and says users should only pay for tokens — matchaman11 · 2026-07-21
- Shared agent workspaces fail in a fixed order, from stale reads to zombie writes — mrvladp · 2026-07-21
- pi 0.81.0 adds first-class integration with llama.cpp server — huggingface · 2026-07-21