ZeroClaw Logo

ZeroClaw

零開銷,零妥協。

別的 AI Agent 動輒吃掉 1GB 記憶體,這個只要 4MB。單一 Rust 二進位,10ms 冷啟動,30+ 頻道,接任何模型。10 美元的樹莓派或正式環境雲端伺服器都能跑——不要 Node.js,不要 Python,不妥協。

快速上手

BETA
# macOS via Homebrew. One command. 🦀
$ brew install zeroclaw

Single Rust binary. No runtime dependencies.

核心能力

memory

極致輕量

Release 建置記憶體佔用不到 5MB,比 OpenClaw 小 99%。10 美元的硬體就能跑——ARM、x86、RISC-V 全支援。

chat_bubble

全頻道覆蓋

Telegram、Discord、Slack、WhatsApp、Signal、iMessage、Matrix、IRC、Lark、DingTalk、QQ、Nostr、Email 等等。

database

混合記憶

基於 SQLite 的向量 + 關鍵字混合搜尋。零外部相依性——不需要 Pinecone,不需要 Elasticsearch,不需要 LangChain。

swap_horiz

隨意替換

每個子系統都是 trait——Provider、Channel、Tool、Memory、Tunnel。改一行設定就能換實作,零程式碼改動。

shield

安全優先

閘道配對、嚴格沙箱、明確白名單、工作區隔離、加密金鑰。頻道預設拒絕策略。

bolt

極速冷啟動

單一二進位 Rust 執行時期,日常操作近乎瞬間啟動。同等硬體上比 OpenClaw 快 400 倍。

用戶怎麼說

@rustacean_dev avatar

"ZeroClaw uses 4MB of RAM. OpenClaw uses over 1GB. That's not an optimization — that's a completely different class of software. Running it on my Raspberry Pi Zero with zero issues."

@rustacean_dev
@devops_mike avatar

"Deployed ZeroClaw on a $10 Orange Pi. It handles Telegram + Discord + Slack channels simultaneously with <5MB memory. My OpenClaw setup needed a $599 Mac mini for the same thing."

@devops_mike
@sarahcodes avatar

"The trait-driven architecture is brilliant. Swapped from OpenAI to a local Ollama model by changing one line in config.toml. No code changes, no recompile. Just works."

@sarahcodes
@cloud_native_tom avatar

"Cold start in under 10ms. I measured it. My Node.js agent takes 8 seconds. ZeroClaw's single binary approach is exactly what edge deployment needs."

@cloud_native_tom
@security_emma avatar

"Finally an AI agent that takes security seriously. Gateway pairing, deny-by-default allowlists, workspace scoping, symlink escape detection. This is how it should be done."

@security_emma
@embedded_chris avatar

"Running ZeroClaw on ARM64, x86, and even RISC-V boards. Same binary workflow everywhere. The portable architecture is a game changer for IoT agent deployments."

@embedded_chris
@rustacean_dev avatar

"ZeroClaw uses 4MB of RAM. OpenClaw uses over 1GB. That's not an optimization — that's a completely different class of software. Running it on my Raspberry Pi Zero with zero issues."

@rustacean_dev
@devops_mike avatar

"Deployed ZeroClaw on a $10 Orange Pi. It handles Telegram + Discord + Slack channels simultaneously with <5MB memory. My OpenClaw setup needed a $599 Mac mini for the same thing."

@devops_mike
@sarahcodes avatar

"The trait-driven architecture is brilliant. Swapped from OpenAI to a local Ollama model by changing one line in config.toml. No code changes, no recompile. Just works."

@sarahcodes
@cloud_native_tom avatar

"Cold start in under 10ms. I measured it. My Node.js agent takes 8 seconds. ZeroClaw's single binary approach is exactly what edge deployment needs."

@cloud_native_tom
@security_emma avatar

"Finally an AI agent that takes security seriously. Gateway pairing, deny-by-default allowlists, workspace scoping, symlink escape detection. This is how it should be done."

@security_emma
@embedded_chris avatar

"Running ZeroClaw on ARM64, x86, and even RISC-V boards. Same binary workflow everywhere. The portable architecture is a game changer for IoT agent deployments."

@embedded_chris
@ivanfioravanti avatar

"The hybrid memory system blew my mind. SQLite vector search + FTS5 keyword search, zero external dependencies. No Pinecone, no Elasticsearch. Just a single file."

@ivanfioravanti
@startup_alex avatar

"Migrated from OpenClaw to ZeroClaw in 20 minutes. 'zeroclaw migrate openclaw --dry-run' showed exactly what would change. Memory, config, everything carried over cleanly."

@startup_alex
@data_lisa avatar

"17 channels supported out of the box. Telegram, Discord, Slack, WhatsApp, Signal, Matrix, IRC, Lark, DingTalk, QQ, Nostr... I've never seen this level of integration coverage."

@data_lisa
@homelab_dan avatar

"ZeroClaw as a systemd service on my home server. Always on, <5MB footprint, handles all my home automation through natural language. This is the future of personal AI."

@homelab_dan
@mobile_chris avatar

"The Docker sandboxed runtime option is perfect for multi-tenant setups. Each user gets their own isolated ZeroClaw instance. Security without complexity."

@mobile_chris
@backend_ninja avatar

"Built by Harvard and MIT students, and it shows. The code quality, the architecture decisions, the documentation — everything is production-grade from day one."

@backend_ninja
@ivanfioravanti avatar

"The hybrid memory system blew my mind. SQLite vector search + FTS5 keyword search, zero external dependencies. No Pinecone, no Elasticsearch. Just a single file."

@ivanfioravanti
@startup_alex avatar

"Migrated from OpenClaw to ZeroClaw in 20 minutes. 'zeroclaw migrate openclaw --dry-run' showed exactly what would change. Memory, config, everything carried over cleanly."

@startup_alex
@data_lisa avatar

"17 channels supported out of the box. Telegram, Discord, Slack, WhatsApp, Signal, Matrix, IRC, Lark, DingTalk, QQ, Nostr... I've never seen this level of integration coverage."

@data_lisa
@homelab_dan avatar

"ZeroClaw as a systemd service on my home server. Always on, <5MB footprint, handles all my home automation through natural language. This is the future of personal AI."

@homelab_dan
@mobile_chris avatar

"The Docker sandboxed runtime option is perfect for multi-tenant setups. Each user gets their own isolated ZeroClaw instance. Security without complexity."

@mobile_chris
@backend_ninja avatar

"Built by Harvard and MIT students, and it shows. The code quality, the architecture decisions, the documentation — everything is production-grade from day one."

@backend_ninja

常見問題

關於 ZeroClaw 的常見問題和解答。

保持關注

獲取新版本、整合和 Rust 驅動的 Agent 基礎設施更新。不發垃圾郵件,隨時退訂。