The hottest project in the AI circle lately is undoubtedly OpenClaw.
Since going open source, GitHub Stars have surged, now nearly breaking 200k, becoming the fastest-growing project in GitHub's history.
Honestly, its features are powerful, but anyone who has used it knows the窒息 memory usage and startup speed are quite painful.
Just as everyone endure狂转 fans while awaiting optimization, a Rust-based rewrite主打极速、轻量： ZeroClaw, quietly open-sourced.
This project, written in Rust, optimizes memory and response speed to the extreme, and with sandbox design, ensures a more secure runtime environment.
Benefiting from its lightweight simplicity, we can deploy it on any server.
Additionally, it has the following features:
• Extreme simplicity: Rust-driven, low overhead,秒启动.
• Native security: Built-in sandbox isolation and pairing mechanism.
• Highly pluggable: Core components are interchangeable.
• Zero vendor lock-in: Wide compatibility with protocols like OpenAI.
If compared with OpenClaw, this guy is a妥妥 performance monster.
ZeroClaw's runtime memory usage is only 7.8MB, nearly 200 times lower than OpenClaw.
In runtime, their startup speeds are completely on different scales. ZeroClaw can be said to open instantly, like calling a simple system command, with no delay.
Plus its tiny size of only 3.4 MB, for friends wanting to deploy Agents on Raspberry Pi or low-config cloud hosts, it's simply an AI神器.
Here is a detailed benchmark comparison chart to give you a feel:
After seeing this data, I相信各位 are already eager.
Without further ado, let's see how to deploy this hardcore ZeroClaw suitable for long-term operation on local servers.
How to Deploy Your First ZeroClaw
1) Environment Preparation
Since ZeroClaw is a pure Rust project, if you haven't installed Rust yet, run this command in the terminal to quickly搞定:
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
2) Compile Version
After the environment is set up, we clone the code and can start compiling and installing.
Here I recommend the Release version because it has the smallest size and fastest speed:
git clone https://github.com/theonlyhennygod/zeroclaw.git
cd zeroclaw
cargo build --release
# Install to system path
cargo install --path . --force
3) Basic Configuration
This project not only installs quickly but also配置极其人性化.
Run this wizard command below:
zeroclaw onboard --interactive
In this process, we only need to do three things:
1. Input the LLM's API Key (supports mainstream models like OpenAI, Anthropic, DeepSeek, OpenRouter).
2. Choose the channel to connect (e.g., Slack, Discord).
3. Security setting: Force set a "pairing code" to prevent strangers from randomly connecting to our Agent.
4) Start Daemon:
zeroclaw daemon
At this point, it runs silently in the background.
We can check its status anytime with the following command:
zeroclaw status
Now, you have a fully functional AI assistant on standby 24/7.
AI Role Definition
Besides basic conversation功能, what惊喜 me most about ZeroClaw is its support for AIEOS (Artificial Intelligence Entity Object Standard).
Simply put, AIEOS is a standard to avoid AI becoming a阅后即焚 tool, focusing on evolving AI into a数字生命实体 with memory, personality, and growth.
Previously, adjusting AI's persona usually involved writing essays in the Prompt.
But ZeroClaw定制izes the AI's soul from the底层 via JSON files. Utilizing the AIEOS standardized framework, it tries to solve the identity misplacement problem of AI agents.
It拆解s the AI's behavior blueprint into the following standardized dimensions:
Identity: Name, background, origin.
Psychology: Cognitive weights, MBTI personality, moral准则.
Linguistics: Text style, catchphrases.
Motivations: Core drivers, short and long-term goals.
For example, if we want to create a character like Elara Vance, an illustrator and translator.
Just import the corresponding AIEOS persona package, and ZeroClaw immediately变成 a creative partner.
Besides helping us精准 translate Chinese blogs into idiomatic English, it can also构思 an artistic配图 for the article.
Of course, enabling AIEOS is also simple. After setting up identity.json, add the following几行配置:
[identity]
format = "aieos"
aieos_path = "identity.json"
This design's portability is very strong. After carefully调教 the cyber soul, we can pack it anytime and migrate to any ecosystem supporting AIEOS standards. The AI's personality and memory are持久化 saved.
OpenClaw and ZeroClaw, How to Choose?
Seeing here, many friends might纠结 between OpenClaw and ZeroClaw, which one to choose?
As an OpenClaw user who已入手 a Mac mini M4 and ran it for a while, my suggestion is that both can be used, but depends on the场景.
Scenario A: Value家庭中枢与创意辅助, choose OpenClaw
If you want to use Mac Mini as a本地服务器,注重与电视交互体验, or want to control it via voice, play with Live Canvas, OpenClaw does indeed do well in human-computer interaction.
Scenario B: 自动化流水线与服务器运维, choose ZeroClaw
If the需求 is to daily定时抓取 blogs,监控服务器日志, or deploy on低配云服务器.
Then, ZeroClaw is undoubtedly the最优选.
Its极低的资源占用 can大幅减少服务器资源 waste. Using the saved memory to run more other business, isn't it香?
写在最后
Currently, we are逐步迈向 Agentic AI时代.
The future internet入口 will no longer be isolated APPs, but countless数字员工 like ZeroClaw,极致精简且无处不在.
They will run on phones, computers, and even every edge node, pushing AI from expensive cloud算力中心 to本地优先的端侧落地.
The appearance of ZeroClaw marks Agent基础设施 starting to对齐 with industrial standards.
It not only抹平s the performance gap with Rust's rigor but also provides the possibility of批量部署 for一人企业.
As more developers投下重注, the signal is very明确.
The second half of AI is no longer about对话的文采, but about executing with最低成本 and最高效率.
GitHub project address: https://github.com/theonlyhennygod/zeroclaw
Today's分享 ends here. Thanks for reading, see you next time, Respect!