Claw Products

The original open-source personal AI agent platform, formerly known as Clawdbot. Created by Peter Steinberger in November 2025, it became the fastest-growing GitHub repo in history, surpassing React to become the most-starred software project. A self-hosted, model-agnostic autonomous agent that integrates with messaging platforms like WhatsApp, Telegram, Slack, and Discord. Now maintained by an open-source foundation after Steinberger joined OpenAI.

Cloud-native OpenClaw deployment built into Moonshot AI's consumer platform kimi.com. Offers one-click agent deployment powered by Kimi K2.5 Thinking model, with 5,000+ ClawHub skills, 40GB cloud storage, and Pro-Grade Search for real-time information retrieval. Currently in beta for Allegretto members and above.

Fully managed cloud AI agent platform by MiniMax, launched February 25, 2026. Powered by MiniMax M2.5 (229B parameter MoE model), it offers one-click deployment, 10,000+ pre-configured Experts, and multi-platform messaging integration. Claims Claude 4.6-level reasoning at roughly 10% of the cost.

Cloud SaaS version of OpenClaw launched by ByteDance's Volcano Engine on March 9, 2026. Codenamed 'Lobster', it deeply integrates with the Volcano Ark model platform and Feishu office suite. Supports multiple LLMs including Doubao-Seed-2.0, Kimi 2.5, MiniMax 2.5, and GLM. Designed for enterprise use with 24/7 operation and built-in security scanning.

Xiaomi's mobile-first AI agent built on its proprietary MiMo large model, announced March 6, 2026. A system-level intelligent agent that operates the phone, invoking 50+ core capabilities and controlling 1 billion+ IoT devices in the Mijia ecosystem. Integrates with Xiaomi's 'human-car-home' vision spanning smartphones, EVs, and smart home devices. Currently in invite-only closed beta for Xiaomi 17 series users.

Open-source personal AI assistant by Alibaba's AgentScope team, short for 'Co Personal Agent Workstation'. Emphasizes local-first operation with support for local models via Ollama/llama.cpp, while also offering cloud deployment through Alibaba Cloud. Supports DingTalk, Lark, QQ, Discord, iMessage, and ClawHub skills. Built on AgentScope framework with persistent long-term memory.

Desktop-level AI agent by NetEase Youdao, widely regarded as the 'Chinese OpenClaw'. Launched February 11, 2026 with a polished Electron + React GUI that installs with a double-click. Features Cowork mode for executing tools, manipulating files, and running commands in a sandboxed environment. Supports Telegram, Discord, DingTalk, and Feishu for remote control.

Open-source 'Agent Operating System' built entirely in Rust by RightNow AI. 137K lines of Rust across 14 crates compiled to a single ~32MB binary. Features 7 autonomous 'Hands' that run on schedules, 40 channel adapters, 38 built-in tools, and 16 security systems including WASM sandboxing, Ed25519 signing, and Merkle audit trails. Starts 33x faster and uses 10x less memory than OpenClaw.

Security-focused lightweight alternative to OpenClaw that runs every agent session inside an isolated Linux container. Created by Gavriel Cohen as a direct response to OpenClaw's security concerns. The entire codebase is ~3,900 lines across 15 files, making it fully auditable. Built on Anthropic's Agents SDK with container-based isolation instead of application-level permission checks. Endorsed by Andrej Karpathy.

Performance-focused Rust reimplementation of OpenClaw's core functionality, built from scratch for resource-constrained environments. Compiles to a 3-5MB binary that uses <5MB RAM and starts in <10ms. Supports 30+ messaging channels, 22+ LLM providers, and can directly read OpenClaw configuration for migration. The lean alternative for VPS, home lab, or Raspberry Pi deployments.

Ultra-lightweight AI agent written in Go by Sipeed, designed to run on sub-$10 embedded hardware with <10MB memory. Targets IoT and edge devices including RISC-V boards, ARM routers, and IP cameras. Pairs with PicoLM (1B parameter model) for fully offline operation. The only Claw agent that officially supports RISC-V architecture. 95% of the code was generated by AI through self-bootstrapping.

The smallest fully autonomous AI agent infrastructure in the ecosystem, written entirely in Zig. Compiles to a 678KB static binary that boots in under 2 milliseconds using ~1MB RAM. Features zero-cost abstractions via Zig vtable interfaces, 22+ LLM providers, 17 messaging channels, and hardware peripheral support for Arduino and Raspberry Pi. Most thoroughly tested in the family with 2,000+ tests.

Multi-agent team platform where specialized AI agents (coder, writer, researcher) collaborate through chain execution and fan-out delegation. Features isolated workspaces per agent to prevent context bleed, a live TUI dashboard for monitoring, and persistent team chatrooms for inter-agent communication. 99.3% smaller than OpenClaw. Built on Claude Code CLI and Codex CLI.

Security-first AI agent runtime by NEAR AI, built in Rust and deployed inside encrypted Trusted Execution Environments (TEEs). Credentials live in an encrypted vault injected at the network boundary - the AI never sees raw values. Every tool runs in its own WebAssembly sandbox with no filesystem access. Announced at NEARCON 2026, it addresses OpenClaw's credential exposure risks by design.

Easy-mode runtime and UI layer built on top of OpenClaw, designed to make AI agents accessible to non-programmers. Available as a desktop app (Mac & Windows), Chrome extension, and managed cloud service. Users interact through natural-language rules and feedback instead of code. Features hot reload for configuration changes, token usage tracking, and speech-to-text integration.

Singapore-based managed hosting platform for OpenClaw, founded in 2024. Provides dedicated VPS instances with OpenClaw pre-configured, supporting 94 curated AI models from 17+ providers. Deploy an AI assistant in under 60 seconds with automatic updates, zero downtime, and 99.975% uptime. Data centers in 6 global regions including US, Europe, Asia Pacific, and Japan.

Managed OpenClaw orchestrator running on Cloudflare's edge network, built on the Moltworker architecture. Automatically provisions secure runtime environments, object storage, and databases for each agent with zero cold starts and global low latency. Features built-in monetization support via Stripe and UZS payment systems. API keys are injected via encrypted secrets and never stored.

Ultra-lightweight personal AI assistant from HKU's Data Intelligence Lab, delivering 90%+ of OpenClaw's core functionality in just ~4,000 lines of Python. Model-agnostic with 11+ LLM providers and 8+ messaging platforms including Telegram, Discord, WhatsApp, and Chinese platforms like QQ, DingTalk, and Feishu. Features stateful graph-based long-term memory. Installable via pip or uv.