How can a number of brokers work collectively like an elite staff to autonomously divide duties, talk successfully, and collaborate seamlessly?
The openJiuwen neighborhood has launched the newest model of JiuwenClaw. This provides help for AgentTeam, a multi-agent collaboration characteristic. The subsequent leap past harness engineering is adjustment engineering.
In detailed testing, this staff collaboration mechanism confirmed outstanding stability. Staff members have clear roles, collaborate autonomously with seamless coordination, and no human intervention is required all through the workflow.
How hardcore are you, actually?
You’ll be able to autonomously assemble a staff of “well-trained” brokers that may create a sturdy and logically rigorous 200-page technical PPT in underneath 20 minutes.
Challenge hyperlink: https://github.com/openJiuwen-ai/jiuwenclaw
Actual check of JiuwenClaw “adjustment engineering”
Need deep insights with out doing something? Create a 200-page content-rich PPT in underneath 20 minutes.
For our trial, we carried out a deep dive into OpenClaw know-how and requested them to investigate it throughout 10 key points. We assigned a devoted agent to deal with every facet. Every agent was liable for creating 20 PPT slides underneath a unified theme. Ultimately, the ten units of slides had been mixed into an entire 200-page technical presentation.
The entire course of took lower than 20 minutes. The ensuing PPT was detailed, logically structured, and extremely environment friendly.
Technical Breakdown: Three Core Options of JiuwenClaw AgentTeam
AgentTeam’s core design philosophy is simple: it simulates how real-world groups collaborate.
- The lead agent is liable for necessities evaluation, staff constructing, and process planning.
- A number of Teammate brokers request duties, carry out them independently, report outcomes, and collaborate via a shared workspace.
- Throughout execution, main milestones require management approval and catastrophe restoration is computerized.
1. Hierarchical autonomous collaboration: Leaders coordinate intelligently, teammates execute autonomously
JiuwenClaw AgentTeam delegates this accountability to the chief agent itself.
What leaders do:
- Dynamically construct groups: Dynamically assign roles and members primarily based on targets. Should you want extra assist throughout the run, you may add or take away members on the fly.
- Plan your duties: Divide your targets into particular duties and set up dependencies (e.g. “Evaluation can solely start after knowledge assortment is full”).
- Assign and monitor: After you create a process, observe its progress in real-time, together with who requested what, who accomplished what, and who encountered any points, and alter accordingly.
What your teammates will do:
- Proactively declare duties: Browse your process board and declare duties that match your capabilities.
- Run independently: Full your work inside your personal workspace.
- Report outcomes: Replace standing and notify leaders and different dependents.
Staff members drive core workflows via collaboration on duties. This implies requesting, executing, finishing, and unblocking downstream duties, discussing plans, negotiating priorities, reporting points, requesting help, and extra.
Each channels run in parallel and process dependencies are routinely managed moderately than simply mechanical distribution and aggregation.
2. Staff Workspace: Shared staff file area
JiuwenClaw AgentTeam solves this drawback with staff workspaces, that are team-level shared file areas which can be transparently accessible to all members. Every teammate’s working listing routinely mounts a shared path pointing to the identical staff workspace.
3. Full lifecycle administration: from plan approval to automated catastrophe restoration
3.1 Chief Approval
AgentTeam supplies a two-tier authorization mechanism.
- plan mode: For essential duties, teammates first submit an execution plan for chief approval.
- Software approval: If teammates must carry out delicate operations (reminiscent of deleting information, calling exterior APIs, or altering sharing settings), they want chief approval.
3.2 Occasions‑drive mechanism
AgentTeam mitigates this with an event-driven mechanism, utilizing each exterior and inside occasions.
- Exterior occasions: Occasions are triggered when there’s a significant change, reminiscent of a change within the state of a process, a change within the lifecycle of a member, or a message between members.
- Inner occasions: Self-checking occasions (mailbox polls, process board polls) generated by the framework function a security web.
When an occasion is triggered, the related brokers are routinely launched (e.g., an idle teammate requests a process, a frontrunner reassigns a process that has timed out).
3.3 Persistent Staff
Enabling persistent mode means that you can persist your staff between periods. The subsequent time you want your staff, you may restore it with one click on. Create a brand new session area, restart your staff members, and also you’re able to go with out having to rebuild your staff from scratch.
3.4 Staff monitor
TeamMonitor supplies observability in two dimensions.
- Question API: You’ll be able to examine the standing of your staff at any time, together with staff info, member standing, process progress, and extra.
- Occasion stream: Subscribe to your staff’s occasions in actual time. All occasions, reminiscent of process completion, member state adjustments, sending and receiving messages, may be consumed one after the other via an asynchronous iterator. You’ll be able to construct dashboards, logging methods, and set off exterior workflows from these occasions. Each step of your staff’s operations is traceable and auditable.
Core basis: openJiuwen AgentTeam structure
AgentTeam’s core technical ideas may be summarized in three factors:
- Constant collaboration with shared process lists: All members share the identical dynamic process checklist. Every agent requests and performs duties independently primarily based on staff targets, process definitions, and distinctive capabilities, guaranteeing pure info consistency.
- Twin-drive mannequin of messages and duties: Members drive the core workflow via process transitions whereas constantly discussing and negotiating via message channels exterior to the duty system, protecting the whole lot from structured execution to unstructured communication.
- Position and gear engineering: Position insurance policies outline the norms of habits and decision-making boundaries for leaders and teammates inside a staff. TeamTools supplies sure coordination capabilities for staff members. Roles decide what it is best to do, and instruments decide what you are able to do.
About Kumonzume
JiuwenClaw is a “Claw” agent developed on the openJiuwen open supply neighborhood. Natively helps multi-agent collaboration and agent self-evolution. The core design philosophy is easy. “Perceive what you need and evolve autonomously.”
In addition to AgentTeam, JiuwenClaw can also be very simple to put in and deploy, getting up and operating with a single command. For the quickstart, see https://github.com/openJiuwen-ai/jiuwenclaw/blob/develop/docs/en/Quickstart.md.
Moreover, JiuwenClaw affords a number of benefits in autonomous process planning, self-evolution, context compression and offloading, browser operation, and general “lobster-like” processing.
- Autonomous process administration: At all times prepared: JiuwenClaw has a process planning mode, which is principally an AI to-do checklist. Customers can dynamically droop, add, or change duties at any time.
- self‑Evolving expertise: Proactively report these execution errors and suggestions, analyze root causes, and generate focused enchancment suggestions. After that, an evolution approval window will pop as much as the consumer. All updates will probably be made upon your request.
- Context compression and offloading: Successfully cut back prices by managing context size.
- layered reminiscence: Realizes long-term storage and clever search of situations and operation tracks.
- Browser operations: Robotically entry profile info reminiscent of cookies and native cache, seamlessly taking up your browser setting.
About OfficeClaw
OfficeClaw, an enterprise-grade model constructed on the Harness engineering basis, seamlessly integrates process planning, multi-agent collaboration, device invocation, and safety governance into Huawei Cloud AgentArts to enhance the success charge of advanced workplace duties.
Be part of the neighborhood and discover openJiuwen
openJiuwen obtain hyperlink
JiuwenClaw obtain hyperlink
- JiuwenClaw on GitHub: https://github.com/openJiuwen-ai/jiuwenclaw
- JiuwenClaw for AtomGit: https://gitcode.com/openJiuwen/jiuwenclaw
- AgentArts on Huawei Cloud: https://www.huaweicloud.com/product/agentarts
- OfficeClaw on Huawei Cloud: https://www.huaweicloud.com/product/agentarts/officeclaw.html
Notice: Due to OpenJiuwen staff for assets, photographs, movies, and different particulars.
Michal Sutter is an information science knowledgeable with a grasp’s diploma in knowledge science from the College of Padova. With a robust basis in statistical evaluation, machine studying, and knowledge engineering, Michal excels at remodeling advanced datasets into actionable insights.
