Indexable Memory (QMD)

QMD (Query Markdown Documents) gives your Claw a persistent, searchable memory that lives entirely on your machine. Notes, research, meeting summaries, project docs — all indexed locally with hybrid search and available to your Claw in any future conversation.

The problem

By default, your Claw's memory is limited to what's in the current conversation window. You can manually save files, but finding the right information later requires knowing exactly where you put it. QMD solves this: save anything to memory, and your Claw can retrieve it later using natural language — without re-reading every file.

How QMD works

QMD is a local-first memory backend that combines three retrieval methods and re-ranks results:

• BM25 — keyword search, fast and precise for exact terms
• Vector search — semantic similarity, finds conceptually related content even when wording differs
• Reranking — scores results by relevance before returning them to your Claw

Everything stays on your machine. No documents are uploaded to any external service.

Steps

npx clawhub@latest install qmd-memory

Set my memory backend to QMD.

[memory]
backend = "qmd"

Save this to memory: "QMD memory is working correctly. Today is the first time I tested it."

What do you remember about QMD memory?

Save a summary of what we just discussed to memory under the projects collection. Label it "API caching decision — May 2026".

After every research session, save a summary of what you found to the intelligence collection.

At the end of each conversation, save any decisions we made to the workspace collection.

What to expect

Once QMD is running, your Claw builds a growing knowledge base it can draw on across all your conversations. You'll notice:

• It remembers context from previous sessions without you repeating yourself
• Research compounds — findings from one conversation inform the next
• Decisions are retrievable: "What did we decide about the auth approach?" actually works

Practical patterns

Save at the end of every session:

Before we wrap up, save a summary of everything we decided today to the workspace collection.

Use the intelligence collection for research:

Research the top 3 competitors for [product]. Save your findings to the intelligence collection tagged "competitor research May 2026".

Recall before starting new work:

Before we start, check memory for anything relevant to [topic].

Build a daily log:

At the end of the day, save a brief log of what I worked on to the daily-logs collection.

Tips

Save summaries, not transcripts. A 200-word summary of a conversation is more useful than the full 4,000-word log. Ask your Claw to summarise before saving.

Tag entries with dates and topics. "Competitor research — May 2026" is much easier to retrieve than "research notes".

QMD replaces G-Brain for most workflows. If you were using G-Brain for knowledge base features, QMD covers the same use cases without cloud dependencies or per-query costs.