This chapter is a step-by-step tutorial on the installation and usage of an essential tools of Openclaw, QMD.
root# useradd -m -s /usr/bin/bash claw_team
root# passwd claw_team
New password: W**T**
Retype new password: W**T**
passwd: password updated successfully
root# usermod -aG sudo claw_team
root# su - claw_team
To run a command as administrator (user "root"), use "sudo <command>".
See "man sudo_root" for details.
# From 'root' switches to 'claw_team'
claw_team$
# Download and install nvm:
claw_team$ curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.4/install.sh | bash
# in lieu of restarting the shell
claw_team$ \. "$HOME/.nvm/nvm.sh"
# Download and install Node.js:
claw_team$ nvm install 24
# Verify npm version:
claw_team$ npm --version
11.9.0
claw_team$ node --version
v24.14.0
# Install pnpm
claw_team$ curl -fsSL https://get.pnpm.io/install.sh | sh -
==> Downloading pnpm binaries 10.32.0
...
To start using pnpm, run:
source /home/claw_team/.bashrc
claw_team$ source /home/claw_team/.bashrc
claw_team$ which pnpm
/home/claw_team/.local/share/pnpm/pnpm
claw_team$ /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
==> Checking for `sudo` access (which may request your password)...
==> This script will install:
/home/linuxbrew/.linuxbrew/bin/brew
/home/linuxbrew/.linuxbrew/share/doc/homebrew
...
==> Next steps:
- Run these commands in your terminal to add Homebrew to your PATH:
echo >> /home/claw_team/.bashrc
echo 'eval "$(/home/linuxbrew/.linuxbrew/bin/brew shellenv bash)"' >>
/home/claw_team/.bashrc
eval "$(/home/linuxbrew/.linuxbrew/bin/brew shellenv bash)"
- Install Homebrew's dependencies if you have sudo access:
sudo apt-get install build-essential
For more information, see:
https://docs.brew.sh/Homebrew-on-Linux
==> Running `brew cleanup gcc`...
Disable this behaviour by setting `HOMEBREW_NO_INSTALL_CLEANUP=1`.
Hide these hints with `HOMEBREW_NO_ENV_HINTS=1` (see `man brew`).
claw_team$ echo >> /home/claw_team/.bashrc
claw_team$ echo 'eval "$(/home/linuxbrew/.linuxbrew/bin/brew shellenv bash)"' >> /home/claw_team/.bashrc
claw_team$ eval "$(/home/linuxbrew/.linuxbrew/bin/brew shellenv bash)"
claw_team$ sudo apt-get install build-essential
claw_team$ brew install gcc
==> Fetching downloads for: gcc
...
We followed the instructon of "3 Essential Tools for OpenClaw", but didn't use ClaudeCode.
Additionally, notice that we use an alibaba ECS located in Virginia, US, so that we don't need to handle the GFW issue.
Set up QMD as the memory backend for my OpenClaw agent.
Follow the official docs here:
https://docs.openclaw.ai/concepts/memory#qmd-backend-experimental
Make sure to:
1. Install the QMD CLI
2. Install SQLite with extension support if needed
(macOS: brew install sqlite)
3. Configure memory.backend = "qmd" in my openclaw.json
4. Add my workspace memory files as a QMD collection
5. Run the initial embed so models are downloaded and
the index is built
6. Verify it works by running a test query
root# su - claw_team
claw_team$ pwd
/home/claw_team/.openclaw
claw_team$ npm --version
11.9.0
claw_team$ node --version
v24.14.0
claw_team$ npm install -g @tobilu/qmd
npm warn deprecated prebuild-install@7.1.3: No longer maintained. Please contact the author of the relevant native addon; alternatives are available.
⠼
added 259 packages in 35s
claw_team$ brew install sqlite
==> Auto-updating Homebrew...
Adjust how often this is run with `$HOMEBREW_AUTO_UPDATE_SECS` or disable with
`$HOMEBREW_NO_AUTO_UPDATE=1`. Hide these hints with `$HOMEBREW_NO_ENV_HINTS=1` (see `man brew`).
...
==> Running `brew cleanup sqlite`...
Disable this behaviour by setting `HOMEBREW_NO_INSTALL_CLEANUP=1`.
Hide these hints with `HOMEBREW_NO_ENV_HINTS=1` (see `man brew`).
claw_team$ which sqlite3
/home/linuxbrew/.linuxbrew/bin/sqlite3
claw_team$ sqlite3 --version
3.52.0 2026-03-06 16:01:44 557aeb43869d3585137b17690cb3b64f7de6921774daae9e56403c3717dceab6 (64-bit)
# Create a tempoerary collection
claw_team$ qmd collection add ~/qmd_notes --name notes
Creating collection 'notes'...
Collection: /home/claw_team/qmd_notes (**/*.md)
No files found matching pattern.
Indexed: 0 new, 0 updated, 0 unchanged, 0 removed
✓ Collection 'notes' created successfully
# Add context to help with search results, each piece of context will be returned
# when matching sub documents are returned.
# This works as a tree. This is the key feature of QMD as it allows LLMs
# to make much better contextual choices when selecting documents.
# Don't sleep on it!
claw_team$ qmd context add qmd://notes "Personal notes and ideas"
✓ Added context for: qmd://notes/ (collection root)
Context: Personal notes and ideas
# Generate embeddings for semantic search
claw_team$ qmd embed
✓ All content hashes already have embeddings.
# Run the search for the first time, expect to find nothing.
# But it will automatically download and install a tiny LLM for reranking,
# tobil/qmd-query-expansion-1.7B, that runs on CPU.
#
claw_team$ qmd query "quarterly planning process"
Expanding query...QMD Warning: no GPU acceleration, running on CPU (slow). Run 'qmd status' for details.
Downloading to ~/.cache/qmd/models
✔ hf_tobil_qmd-query-expansion-1.7B-q4_k_m.gguf downloaded 1.28GB in 4m
Downloaded to ~/.cache/qmd/models/hf_tobil_qmd-query-expansion-1.7B-q4_k_m.gguf
(290.6s)
├─ quarterly planning process
├─ lex: definition of quarterly planning
├─ vec: definition of quarterly planning and its significance
├─ vec: importance of setting quarterly goals for businesses
└─ hyde: The topic of quarterly planning process covers debates surrounding th...
Searching 5 queries...
No results found.
claw_team$ tree ~/.cache/qmd/models
/home/claw_team/.cache/qmd/models
└── hf_tobil_qmd-query-expansion-1.7B-q4_k_m.gguf
1 directory, 1 file
# Add a temporary MD file for testing.
claw_team$ mkdir -p ~/qmd_notes
claw_team$ cd qmd_notes/
claw_team$ curl https://raw.githubusercontent.com/cft0808/edict/main/README.md -o 三省六部.md
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 29820 100 29820 0 0 284k 0 --:--:-- --:--:-- --:--:-- 285k
# Update the QMD index
#
claw_team$ qmd update
Updating 1 collection(s)...
[1/1] notes (**/*.md)
Collection: /home/claw_team/qmd_notes (**/*.md)
Indexing: 1/1
Indexed: 1 new, 0 updated, 0 unchanged, 0 removed
✓ All collections updated.
Run 'qmd embed' to update embeddings (1 unique hashes need vectors)
# QMD will automatically download and install another LLM that runs on CPU, to rerank the hybrid search results.
# This time, QMD downloaded, qwen3-reranker-0.6b-q8_0.gguf,
#
claw_team$ qmd query "尚书省"
Warning: 1 documents (100%) need embeddings. Run 'qmd embed' for better results.
Expanding query...QMD Warning: no GPU acceleration, running on CPU (slow). Run 'qmd status' for details.
(8.4s)
├─ 尚书省
├─ lex: 汉朝尚书省
├─ lex: 古代尚书中书府
├─ vec: 汉朝尚书省
├─ vec: 古代尚书中书府
└─ hyde: The topic of 尚书省 covers 古代尚书中书府. Proper implementation follows establ...
Searching 6 queries...
Downloading to ~/.cache/qmd/models
✔ hf_ggml-org_qwen3-reranker-0.6b-q8_0.gguf downloaded 639.15MB in 1m
Downloaded to ~/.cache/qmd/models/hf_ggml-org_qwen3-reranker-0.6b-q8_0.gguf
(91.4s)
qmd://notes/三省六部.md:8 #6bf8ae
Title: 🎬 Demo
Context: Personal notes and ideas
Score: 88%
@@ -7,4 @@ (6 before, 100 after)
<p align="center">
<sub>12 个 AI Agent（11 个业务角色 + 1 个兼容角色）组成三省六部：太子分拣、中书省规划、门下省审核封驳、尚书省派发、六部+吏部并行执行。<br>比 CrewAI 多一层<b>制度性审核</b>，比 AutoGen 多一个<b>实时看板</b>。</sub>
</p>
# So far, QMD has downloaded 3 LLMs for reranking.
#
# claw_team$ ls -la ~/.cache/qmd/models/
total 2197456
drwxrwxr-x 2 claw_team claw_team 4096 Mar 16 22:28 .
drwxrwxr-x 3 claw_team claw_team 4096 Mar 16 23:28 ..
-rw-rw-r-- 1 claw_team claw_team 328576992 Mar 16 22:28 hf_ggml-org_embeddinggemma-300M-Q8_0.gguf
-rw-rw-r-- 1 claw_team claw_team 639153184 Mar 16 22:26 hf_ggml-org_qwen3-reranker-0.6b-q8_0.gguf
-rw-rw-r-- 1 claw_team claw_team 1282438912 Mar 16 21:53 hf_tobil_qmd-query-expansion-1.7B-q4_k_m.gguf
# Remove the temporary collection.
#
claw_team$ qmd collection remove notes
✓ Removed collection 'notes'
Deleted 14 documents
Cleaned up 1 orphaned content hashes
claw_team$ qmd collection add /home/claw_team/.openclaw/workspace --name workspace --mask "**/*.md"
Creating collection 'workspace'...
Collection: /home/claw_team/.openclaw/workspace (**/*.md)
Indexing: 7/7 ETA: 0s
Indexed: 7 new, 0 updated, 0 unchanged, 0 removed
Run 'qmd embed' to update embeddings (7 unique hashes need vectors)
✓ Collection 'workspace' created successfully
claw_team$ qmd context add /home/claw_team/.openclaw/workspace "Agent workspace - memory files, daily logs, projects, skills"
✓ Added context for: qmd://workspace/
Context: Agent workspace - memory files, daily logs, projects, skills
# Generate embeddings for semantic search
#
claw_team$ qmd embed
✓ All content hashes already have embeddings.
# Update the QMD index
#
claw_team$ qmd update
Updating 1 collection(s)...
claw_team$ qmd query "Boundaries"
Warning: 7 documents (100%) need embeddings. Run 'qmd embed' for better results.
Expanding query...QMD Warning: no GPU acceleration, running on CPU (slow). Run 'qmd status' for details.
(6.9s)
├─ Boundaries
└─ hyde: Boundaries is an important concept that relates to limit scenarios. I...
Searching 2 queries...
Embedding 2 queries... (740ms)
Reranking 2 chunks... (5.6s)
qmd://workspace/bootstrap.md:35 #5c4cb8
Title: BOOTSTRAP.md - Hello, World
Context: Agent workspace - memory files, daily logs, projects, skills
Score: 88%
@@ -34,4 @@ (33 before, 19 after)
- How they want you to behave
- Any boundaries or preferences
Write it down. Make it real.
qmd://workspace/soul.md:17 #95c467
Title: SOUL.md - Who You Are
Context: Agent workspace - memory files, daily logs, projects, skills
Score: 53%
@@ -16,4 @@ (15 before, 18 after)
## Boundaries
- Private things stay private. Period.
Add memory section to "/home/claw_team/.openclaw/openclaw.json"
"agents": {
...
},
"memory": {
"backend": "qmd",
"qmd": {
"searchMode": "search",
"includeDefaultMemory": true,
"sessions": {
"enabled": true
},
"paths": [
{ "name": "workspace", "path": "/home/claw_team/.openclaw/workspace", "pattern": "**/*.md" }
]
}
}
In the Openclaw chat channel, type in a query,
请使用 QMD 工具，帮我查找之前笔记中关于 "Boundaries" 的记录
Or, without mentioning QMD,
请帮我查找之前笔记中关于 "Boundaries" 的记录
The Openclaw will use QMD to do both keyword search and semantic search in the /workspace collection,
and merge the two result lists using the CPU-based reranking LLM.
Following is the hybrid search results,
