HiClaw - Alibaba Cloud's open-source multi-agent team collaboration system

What is HiClaw?

HiClaw is an open-source agent-based team collaboration system from Alibaba, positioned as a "Team version of OpenClaw." The system introduces a Manager Agent as an AI steward, automatically coordinating multiple Worker Agents to complete complex tasks. HiClaw's core highlights include: Workers do not hold real credentials (only Consumer Tokens), ensuring secure isolation and preventing leaks; a built-in Matrix server allows for real-time monitoring and intervention via mobile phone; conversational Worker creation, automatic task assignment, and progress monitoring. The system is suitable for independent developers and one-person companies to quickly build AI employee teams, enabling "one person to command multiple agents in collaboration."

HiClaw's main functions

• Manager-Agent CoordinationThe Manager Agent acts as an AI steward, automatically creating Workers, assigning tasks, and monitoring execution progress through natural language dialogue, eliminating the need for manual configuration of each Agent.
• Multi-agent collaborationIt supports the creation of multiple Worker Agents by role (such as front-end, back-end, and product). Each Worker has an independent skill set and memory space, avoiding task mixing and context pollution.
• Security Credential ManagementWorkers only hold the Consumer Token issued by Higress, while sensitive credentials such as the real API Key and GitHub PAT are stored uniformly in the AI Gateway, ensuring that even if a Worker is compromised, the core credentials cannot be leaked.
• Built-in IM systemIt integrates the Tuwunel Matrix server and Element Web client, eliminating the need to connect to Lark or DingTalk robots. It supports real-time communication between browsers and mobile devices out of the box.
• Human supervision interventionAll Agent communication occurs in the Matrix group chat room. Administrators can observe the conversation flow at any time and directly @ any Agent to intervene or make corrections.
• Dynamic skill loadingWorker has a built-in find-skills capability, which can actively search for and install skills.sh community skills, and also supports switching to a private skill library to meet enterprise customization needs.
• File sharing storageUsing MinIO as a centralized file system, intermediate products such as code and documentation are transmitted through file sharing rather than group chat, keeping the conversation context concise.
• One-click deployment and maintenanceA single command can complete the installation of the Higress gateway, Matrix server, MinIO storage, Manager, and web client, supporting in-place upgrades and data persistence.

HiClaw's Technical Principles

• Layered architecture designHiClaw employs a four-layer architecture to achieve separation of duties. The user layer interacts with the system via the Matrix protocol; the coordination layer, managed by the Manager Agent, handles task breakdown and Worker lifecycle management; the execution layer deploys stateless Worker Agent containers; and the gateway layer, managed by Higress AI Gateway, uniformly proxies all external service access and hosts genuine credentials. This layered approach ensures that even if a Worker is attacked, it cannot access sensitive information, while also facilitating independent expansion of each layer's components.
• Security credential transfer mechanismWhen a Worker starts, it obtains a Consumer Token from the Manager. This token only identifies the worker and does not grant actual permissions. When a Worker calls the LLM or GitHub API, the request first passes through the Higress Gateway. The gateway injects the corresponding real API Key based on the token's identity and then forwards the request to the upstream service. This design achieves complete credential isolation: the Manager knows the task content but cannot access the key, and the Worker executes the task but cannot export credentials, thus eliminating the risk of credential leakage from an architectural perspective.
• Anti-frightening and message routingWhen implementing multi-agent collaboration based on Matrix group chat, HiClaw designs a precise wake-up mechanism. Group chat messages do not trigger any agents by default; only when the message content @ a specific agent or contains specific keywords will the corresponding agent call LLM to respond. This avoids the wasteful cost of all agents performing inference for every message, while maintaining the coherence of the group chat context and keeping the communication overhead controllable during multi-person, multi-agent collaboration.
• Tiered storage and state managementThe system employs a dual-storage strategy to separate different types of data. The Matrix server retains only dialogue messages and decision records, serving as the "memory" context for the Agent; MinIO object storage is responsible for storing large intermediate products such as code files, documents, and temporary data. The Worker itself is designed to be stateless, accessing files by mounting shared volumes and being destroyed after the task is completed. This controls the length of the LLM context to reduce costs while enabling the Worker to start, stop, and scale elastically.

HiClaw project address

• Project official websitehttps://higress.ai/hiclaw
• GitHub repositoryhttps://github.com/higress-group/hiclaw

HiClaw Application Scenarios

• Independent developer full-stack developmentIndividual developers can quickly build virtual teams and create multiple workers such as front-end, back-end, and testing for parallel development. The Manager automatically coordinates dependencies, achieving the efficiency of "one person doing the work of a team", which is suitable for independent contracting or side projects.
• One-person company MVP quick validationEntrepreneurs describe product requirements using natural language, and the Manager automatically breaks them down into sub-tasks such as PRD writing, technology selection, core development, and preparation of release materials, which are then assigned to different Workers to execute, completing the entire process from idea to launch within a few days.
• Enterprise Digital Employee PilotEnterprises can deploy HiClaw locally and privately, delegating repetitive tasks (such as data report generation, code review, and document organization) to the Agent team, while human employees focus on decision-making and innovation, gradually exploring a human-machine collaboration model.
• Open source project maintenanceOpen source maintainers create Workers to handle daily tasks, such as automatically responding to issues, generating changelogs, reviewing PRs, and synchronizing document translations. The Manager monitors task status and alerts maintainers to intervene when anomalies occur.