FlowState-QMD: memory tool for Hermes Agent

Local markdown memory server for coding agents that prefetches context before the agent searches

What FlowState-QMD does

FlowState-QMD turns local markdown knowledge into shared project memory for Codex-, Claude-, Cursor- and MCP-style agents. It combines a durable markdown memory store, organized in named collections, with a FlowState anticipatory cache. The cache is written to ~/.cache/qmd/intuition.json so an agent can start from relevant documents instead of deciding to search after the fact. The repository is described as a Hermes 2026 Hackathon entry and is tagged hermes-agent.

Everything runs locally on SQLite, sqlite-vec and local GGUF models through node-llama-cpp. The default interface is an MCP server started with qmd mcp, plus a packaged agent skill. The qmd flow command watches an agent session log and keeps the cache warm, and agents can call fetch_anticipatory_context before a regular search. Results keep file paths, doc IDs, snippets and explain traces. Human-written context overlays are added with qmd context add. It works best on docs, ADRs, RFCs, runbooks, changelogs and migration logs.

Key features

Durable markdown memory in named collections
FlowState anticipatory cache stored at ~/.cache/qmd/intuition.json
MCP server started with qmd mcp, plus a packaged agent skill
qmd flow to watch an agent session log and keep the cache warm
Results with file paths, doc IDs, snippets and explain traces
Local-first operation on SQLite, sqlite-vec and GGUF models

When to use it

Letting a coding agent find the ADR or changelog entry behind a past decision
Answering questions about an incident or rollback from migration notes
Giving several agents the same indexed project memory

Who it is for: Developers who keep project knowledge in markdown and want coding agents to pull relevant context automatically.

How it fits with Hermes Agent

The repository is described as a Hermes 2026 Hackathon entry and carries the hermes-agent topic. It is exposed as an MCP server, while the README itself names Codex, Claude and Cursor-style agents.

How to install FlowState-QMD

These commands are copied from the project's README. Check the repository for the latest steps before you run them.

git clone https://github.com/amanning3390/flowstate-qmd.git
cd flowstate-qmd
bun install
qmd init --target all

Requirements: Bun for the bun install step. Runs locally with SQLite, sqlite-vec and local GGUF models through node-llama-cpp.

Note: It is described as a hackathon entry, and the most recent push was in March 2026.

FAQ

What is FlowState-QMD?

FlowState-QMD is a local memory layer that indexes markdown files and serves them to coding agents over MCP. Its anticipatory cache prefetches likely-relevant context before the agent runs a search.

Does FlowState-QMD work with Hermes Agent?

The repository is described as a Hermes 2026 Hackathon entry and has the hermes-agent topic. It runs as an MCP server, and the README itself names Codex, Claude and Cursor-style agents.

How do I install FlowState-QMD?

Clone the repository, run bun install, then run qmd init --target all and qmd doctor. After that you add collections with qmd collection add, run qmd embed, and start the server with qmd mcp.