How QMD Makes Your AI Credits Go 5x Further
We're transparent about how ClawOps keeps costs low. The secret is QMD, an open-source local search engine that cuts token usage by 80-90%.
“How Can $10/Month in Credits Be Enough?”
It’s the first question people ask. Ten dollars of AI credit sounds thin, until you understand what’s running under the hood.
The answer is QMD, an on-device search engine created by Tobias Lutke. It’s completely open source, it runs entirely on your server, and it’s the single biggest reason your credits go further with ClawOps than anywhere else.
We think you deserve to know exactly how this works.
The Problem: Full-Context Loading
When you ask an AI assistant about a document, the naive approach is to load the entire thing into the LLM’s context window. Your 50-page knowledge base? That’s 40,000+ tokens just to answer one question.
At Claude Sonnet rates, that’s roughly $0.15 per query. Ask 100 questions a month and you’ve burned through $15, just on document lookups. Your $10 credit wouldn’t last two weeks.
This is how most AI assistants work. It’s expensive, wasteful, and unnecessary.
The Fix: Local Search First, LLM Second
QMD indexes your markdown files locally using SQLite and BM25, the same ranking algorithm that powers search engines. When your bot needs to answer a question, it doesn’t send everything to the LLM. Instead:
- Your query hits QMD first, running entirely on your VPS, not the LLM
- QMD returns only the relevant chunks, typically 2-5 snippets of ~800 tokens each
- Only those chunks go to the LLM, instead of the full corpus
The LLM never sees the stuff it doesn’t need. The result: 80-90% fewer tokens per document query.
Real Numbers
| Scenario | Without QMD | With QMD |
|---|---|---|
| Query a 100-page PDF | ~40,000 tokens ($0.12) | ~3,000 tokens ($0.009) |
| 100 document queries/month | $12.00 | $0.90 |
| Monthly savings | $0 | $11.10 (92%) |
This is why $10/month in OpenRouter credit — included with every ClawOps plan — is actually enough for real, daily usage. QMD does the heavy lifting locally. The LLM only sees what it needs to see.
Three Search Modes
QMD isn’t a toy. It’s a serious retrieval engine with three modes:
- BM25 keyword search, instant, handles 90% of queries perfectly
- Vector search, semantic matching using local embeddings when keywords fail
- Hybrid search, combines both with reciprocal rank fusion for maximum recall
All of this runs on your dedicated VPS. No API calls. No extra cost. No data leaving your server.
300+ Models, You Choose
QMD handles the retrieval side. For the AI side, every ClawOps bot connects to OpenRouter, giving you access to 300+ models:
- Anthropic, Claude Opus, Sonnet, Haiku
- OpenAI, GPT-4o, GPT-4o-mini
- Google, Gemini 2.5 Pro, Gemini 2.5 Flash
- DeepSeek, DeepSeek R1, DeepSeek V3
- Meta, Llama 4
- And hundreds more, including free-tier models
We ship with smart defaults, models that balance quality and cost. But you can switch to anything OpenRouter supports. Want Claude Opus for complex reasoning and Gemini Flash for quick lookups? Go for it. It’s your credit, your choice.
Every plan includes $10/month in credits. Need more? Top up with $5, $10, or $20 from your dashboard anytime. How you spend within your balance is entirely up to you.
Credit Where It’s Due
QMD is open source, built by Tobias Lutke. We didn’t invent this, we just made sure every ClawOps bot ships with it pre-configured and indexed.
ClawOps is built entirely on open-source tools like QMD, OpenClaw, Tailwind CSS, Astro, Node.js, and Bun. We take that seriously, 1% of all subscription revenue goes back to these projects every quarter. Read more about that in our OSS commitment post.
Try It Yourself
Every ClawOps bot comes with QMD pre-installed. Your documents get indexed automatically. You don’t have to configure anything.
Or if you want to set up QMD yourself: github.com/tobi/qmd. It’s open source. We’ll even cheer you on.