Unleash self-reasoning CLI beasts to execute code, browse the web, and dominate tasks — across 12+ messaging channels, with voice wake, live canvas A2UI, and multi-agent routing.
ManusClaw is no longer just a chatbot wrapper. It has evolved into a true omnichannel AI platform.
Agents now operate natively across Telegram, Discord, Slack, WhatsApp, Signal, Teams, Matrix, WebChat, and Email with cross-platform routing.
Includes Porcupine wake-word detection (e.g., "Hey Manus") and a continuous talk loop. Command the agent via microphone and get TTS responses.
Agent-to-UI protocol enabled. Agents can now render real-time UI components, charts, tables, and interactive buttons via WebSockets.
Features HMAC-verified incoming webhooks and Gmail Pub/Sub automation. Agents can handle tasks and auto-replies immediately as emails arrive.
A fully independent DAG pipeline operating on a strict PAORR (Plan -> Act -> Observe -> Reflect -> Retry) execution loop.
Analyzes task requirements and generates a PRD (Product Requirement Document). Sets clear, measurable goals.
Plans the software architecture, defines directory trees and the tech stack. Creates a perfect DAG execution plan.
Utilizes the 14-tool arsenal to write code, execute it, and autonomously fix errors without human intervention.
Ensures quality assurance by running rigorous tests and validating that everything is production-ready according to the PRD.
The agent uses these tools to interact with the real world. Soft shapes hiding industrial power.
Stateful shell sessions that remain active in memory. Control the OS natively across tool calls.
Executes code in a fresh, isolated multiprocessing sandbox with a strict 2GB memory limit.
Playwright-driven Chromium automation. Grants the freedom to navigate, click, and extract web data.
New addition! The agent can now read, send, and reply in threads autonomously.
Renders live charts, UI components, and interactive elements directly to connected screens.
SQLite + FTS5 memory integration. Grants the ability to recall past sessions and learn from failures.
The framework's solid logic and robust security systems are represented through these realistic elements.
Tools are auto-penalized on failure. Features seamless failover rotation between LLM providers (OpenAI, Groq, Anthropic).
IdentityGuard stops jailbreaks using 30+ regex patterns. The 3-tier Permission Gate blocks the execution of dangerous commands.
WAL-mode SQLite logging. Every step, tool call, stdout, and error trace is permanently recorded. Zero data loss.
ManusClaw can be deployed in any environment, from heavy backend servers to mobile smartphones.
Wake up the AI shell with the native terminal command manusclaw. Slash commands included.
New feature! Runs an asyncssh server on port 2222, providing remote access via a restricted shell.
Includes Flet Desktop GUI, macOS Menu Bar, and Windows System Tray companion applications.
Supports Docker, SSH remote host, and OpenShell (Linux namespaces) for secure code isolation.
If ManusClaw has empowered your workflow, consider supporting its continued development. Every contribution fuels the next evolution of autonomous AI agents.
If ManusClaw has empowered your workflow, support the project to keep it alive and cover the API testing costs.