03 Core concepts and features
Expert teams: workflow orchestration and process auditing
Expert teams let specialized roles collaborate on a research task. The Lead understands the request, plans the work, and combines results, while members handle search, computation, analysis, writing, or review. Save recurring responsibilities and execution sequences as reusable collaboration patterns.
Use expert team mode
- Open the mode menu at the bottom right of the chat input and select Expert team.
- Review team descriptions and member counts, then choose a team for the task.
- Enter the research goal, provide supporting materials, and describe the deliverables you want.
- After sending, answer follow-up questions or confirm the plan, then review the team's progress and outputs.
You can also open Experts · Skills → Expert teams in the sidebar, view a team's details, and click Summon to start a conversation with it.
Create and manage teams
The list has Official teams and My teams. Use official teams directly, or duplicate one into your own teams before changing its responsibilities.
Click New team and describe the problem, desired outputs, and specialist roles. After creation, use Team editor to manage:
Setting | Purpose |
|---|---|
Team information | Edit the team's name and description and view its change history. |
Members | Configure each member's responsibilities, model, skills, and working requirements. |
Collaboration patterns | Save recurring workflows with stages, parallel members, and reviews. |
For each role, specify the question it owns, the materials it may use, and what it must pass to the next stage. For example, a literature member collects evidence and sources, a computation member provides parameters, logs, and results tables, and a reviewer checks whether conclusions are supported.
Create a collaboration pattern
Define responsibilities, execution order, parallel steps, and human checkpoints using natural language or a drag-and-drop canvas. Different Agents can handle literature research, computation, experimental planning, and reporting. Save the workflow as a task pattern for the Lead to select for new tasks, and add specialists as research needs evolve.
For a systematic literature review, the workflow might include searching, screening, data extraction, evidence synthesis, and delivery, with explicit parallel work, checkpoints, and return paths for missing information. Experimental planning can likewise combine prior research, calculation results, and experimental conditions, with researcher confirmation before execution.
In Collaboration patterns, choose Generate from a sentence or Start from blank, then enter a name and use case. The Lead prioritizes a suitable pattern for the request, or organizes the work based on the task when no pattern fits.
A pattern starts with Lead clarification and planning and ends with Lead synthesis and delivery. Add intermediate stages to suit your research workflow.
- Click Add stage, or an add control on the canvas, and select a member.
- Click a stage to rename it or add members. Members in the same stage can work in parallel.
- Drag members between stages, or use Move left and Move right to reorder stages.
- Click a member to set its requirements for the stage. Advanced settings let you define input materials, output files, read/write scope, and acceptance criteria.
- Use undo, redo, and zoom to review the workflow and check handoffs between stages.
Below the canvas, Questions to confirm before starting stores information needed before each run. The Delivery template defines the final output's structure and format.
Set up reviews and rework
Click the Lead node after a stage and enable Review. After combining that stage's outputs, the Lead runs the configured reviews in order and proceeds once all have passed.
Each review step needs a reviewer, review requirements, and a stage to return to if it fails. Choose a team member or I'll confirm. For member reviews, you can set a maximum rework count. Add, remove, or reorder review steps as needed.
Make review criteria concrete: does every key conclusion have a source, do sample IDs and units match across tables, and are calculation parameters and result files complete? Send rework to the stage responsible so it can rerun and resubmit the work.
Save the collaboration pattern, then save the team. Return to the list to check the pattern's name and workflow before reusing it in team tasks.
View execution progress
Once the team starts, click Execution progress to open the task graph. View task dependencies, parallel assignments, and Lead nodes. Use Locate current task or zoom to find running or waiting stages.
Click a node to open Task details on the right, including the task, planned deliverables, prerequisites, latest progress, and outputs so far. Registered outputs can be opened here.
Lead nodes also show synthesis notes, review steps and status, review comments, and rework history. When a task is Waiting for your review, use the confirmation card in the conversation. If a review fails or reaches the rework limit, read the comments before choosing the next action.
Use View round to switch between current and previous rounds and inspect their tasks, reviews, and outputs. To check a research result, trace the final report to its member tasks, then review the materials, tools, rework, and human confirmations involved.
The team's Change history records edits to its definition; the task graph records how a particular run progressed. Together, they show how the team configuration changed and how a result was produced.