Back to feed
News Story
机器之心
1 sources

JiuwenSwarm Open-Source Unified Workbench Released, Pioneering New Paradigm of Human-Agent Collaboration

JiuwenSwarm, under the Huawei-backed open-source AI Agent platform openJiuwen, has released a unified workbench merging Work and Code modes, along with a new human-agent collaboration paradigm called HITS (Human in the Swarm). This enables humans to collaborate with AI agent teams for office work, programming, and gaming, advancing multi-agent coordination engineering toward human-machine synergy.

SynthePulse Insight · AI deep reading

JiuwenSwarm Unified Workbench: From Multi-Agent Collaboration to a New Paradigm of Human-Machine Symbiosis

Version 1 · 1 source

While the AI community debates Graph Engineering versus Loop Engineering, the Huawei-backed open-source platform openJiuwen has launched the JiuwenSwarm Unified Workbench and introduced a new human-machine collaboration paradigm called HITS. This article deconstructs the core logic: the next leverage point in coordination engineering is not just agent-to-agent collaboration, but how humans step into the swarm.

  • JiuwenSwarm merges Work mode and Code mode into a unified workbench, eliminating the need to switch between writing documents and developing code.
  • Launches the HITS (Human in the Swarm) paradigm, allowing humans to join agent teams for collaborative office work, programming, and even gaming.
  • Office scenario: generates a 200-page high-quality PPT in 20 minutes; wake up the Agent team via Feishu on a high-speed train for remote collaboration.
  • Programming scenario: after the user describes requirements, the system autonomously forms a development team to work in parallel, e.g., recreating the Tank Battle game.
  • Entertainment scenario: humans and AI team up to play Gomoku or Werewolf; AI can act as host or player, supporting private chat.
  • All code is open-source, with one-click installers for Windows, Mac, and HarmonyOS.
Open section navigationFrom Graph Engineering to Coordination Engineering: The Shift in Leverage Points

From Graph Engineering to Coordination Engineering: The Shift in Leverage Points

The AI tech community has recently been discussing Graph Engineering—designing multi-agent organizations as programmable graphs. Meanwhile, another camp believes Loop Engineering (refining a single agent's execution loop) has peaked. Behind the debate lies a consensus: the leverage point is shifting from 'individual agent performance' to 'how multiple agents collaborate.'

The Huawei-backed open-source AI Agent platform openJiuwen proposed Coordination Engineering as early as April, focusing on how to form agent teams, negotiate dynamically, and accumulate collaboration experience. JiuwenSwarm Swarm Agents are the open-source engineering delivery of this paradigm.

Unified Workbench: Seamless Integration of Office and Development

Recently, JiuwenSwarm released a major update: Work mode and Code mode merged into a single unified workbench. Users no longer need to switch between writing materials and developing code; they assign tasks, track progress, and deliver results in one place.

In office scenarios, users can wake up the JiuwenSwarm reporting material team via Feishu on a high-speed train. Tasks execute automatically, and humans can reject outlines, bring in expert agents for consultation, or add beautification agents mid-task. The entire process is a human-agent collaboration.

In cluster mode, after inputting the topic, audience, and style requirements, the system automatically forms a multi-agent team that works in parallel: research, structure, content, and integration. The originally serial process is broken into parallel task chains, generating a 200-page high-quality PPT in 20 minutes.

Programming and Entertainment: New Experiences Under the HITS Paradigm

In programming scenarios, users use Code's cluster mode, and the system forms a small development team: design, frontend, backend, and testing work in parallel, with a Leader coordinating the pace. For example, a user wanting to revisit Tank Battle sees the game run successfully after a few rounds of collaboration; a fitness user describes requirements, and the system autonomously plans tasks and generates a calorie tracking app.

Entertainment scenarios showcase another possibility of HITS: in Gomoku, an agent acts as host, inviting two human players to play, synchronizing the board, maintaining rules, and recommending moves. In Werewolf, humans and AI players team up; AI can reason identities, bluff, vote, and support private chat.

These cases show that agents can not only execute tasks but also coordinate multi-person activities as organizers. Humans no longer stand outside the process but enter the swarm with their identity, reasoning, competing, and delivering together with AI.

Open Ecosystem and Quick Start

All code for JiuwenSwarm is open-source, with one-click installers for Windows, Mac, and HarmonyOS. Users can experience multi-agent tasks after configuring the model service.

The project is hosted on AtomGit and GitHub, allowing the community to freely download, modify, and contribute.

Credibility boundary

This article is based on the official openJiuwen press release published by Machine Heart, which is a product launch promotional material. Performance data (e.g., 200-page PPT in 20 minutes) are official claims and have not been independently verified. Case descriptions are official use scenarios and may have been optimized.

Insight takeaway

The JiuwenSwarm Unified Workbench and HITS paradigm mark the expansion of multi-agent collaboration from inter-agent cooperation to deep human-machine synergy. The leverage point has shifted from individual agent performance to team collaboration, and how humans integrate into the swarm becomes the key issue for the next stage.

Sources for this version

  1. JiuwenSwarm开源统一工作台来了，引领智能体人机协同新范式！机器之心

Primary report

机器之心

Primary source