Remnic

What Remnic does

Remnic keeps agent memory as plain markdown files with YAML frontmatter on your own disk, with no database, cloud dependency or subscription. The files can be read, grepped, edited and version-controlled with ordinary tools. OpenClaw, Claude Code, Codex CLI, Cursor, ChatGPT in developer mode, Hermes, Replit, Pi, omp, Factory Droid and any MCP client read and write the same store, so a preference told to one agent is known to the others.

Remnic watches conversations, distills durable knowledge and injects relevant context back when it is needed. Retrieval is hybrid, combining BM25, vector search and reranking over rebuildable indexes, with graph recall, memory-worth scoring and per-result provenance that can be inspected. Access is through MCP or HTTP. The repository also contains What Helps Me, a private support passport built on Remnic where the owner reviews every card before a helper can see it, created as a Build for Good 2026 entry.

Key features

• Memories stored as markdown files with YAML frontmatter that you can grep, edit and version-control
• One store read and written by Hermes, OpenClaw, Claude Code, Codex CLI, Cursor and any MCP client
• Automatic extraction of durable knowledge from conversations and recall of relevant context
• Hybrid search (BM25, vector and reranking) with graph recall and per-result provenance
• MCP and HTTP access to the memory store

When to use it

• Sharing preferences between Hermes Agent and a coding agent such as Claude Code
• Keeping agent memory in plain files under version control
• Checking why a particular memory was retrieved by reading its provenance

Who it is for: People who run more than one AI agent and want a single local memory they can read and edit as files.

How it fits with Hermes Agent

Hermes is one of the agents that can read and write a Remnic store, and the repository carries the hermes-plugin topic; Remnic itself is a general memory layer.

FAQ