Chat Flows
Chat Flows let you create multi-agent conversations where multiple specialized AI participants collaborate on complex tasks. Think of it as assembling a team of AI experts who each bring different skills to the table.
Overview
In standard chat mode, you talk to one AI model at a time. Chat Flows extend this by introducing multiple AI participants, each with their own:
- Name: A display name for identification
- System prompt: Instructions that define the participant's expertise and personality
- Model: Each participant can use a different AI model
Participants take turns responding to your messages, building on each other's contributions.
Creating a Flow
- Open the model picker
Tap the model selector in the AI Chat screen.
- Switch to Chat Flow tab
Select the "Chat Flow" tab in the model picker.
- Create a new flow
Tap "Create Flow" and give it a descriptive name.
- Add participants
Add AI participants. You can create custom participants or select from existing AI Roles.
Managing Participants
Each flow can have multiple AI participants:
- Add from Roles: Select from your predefined AI Roles to quickly add specialized participants
- Custom participants: Create new participants with custom names, system prompts, and model selections
- Tool Allowlist: Configure which tools each participant can use (max 5), specific to their task
- Reorder: Drag to reorder participants â this determines the response order
- Edit: Modify any participant's name, system prompt, or model at any time
- Remove: Remove participants you no longer need
If participants can use the PRO task_planner tool, they share the same task plan within
the conversation. This enables hand-off workflows where one participant plans and others execute.
Starting a new conversation begins with an empty plan.
When you add a participant from a Role, the role's data is copied. Changes to the original role won't affect existing flow participants.
Context Base Selector
Each participant can configure what message their response is based on. This controls both what the participant "sees" in their context and determines message provenance in the conversation history.
- Follow Previous Message (default): The participant responds to the most recent message in the conversation. This creates a chain where each response builds on the previous one.
- Use Latest User Message: The participant responds directly to your latest input, ignoring any intermediate responses. This enables parallel "fan-out" patterns.
- Use Latest Message From Role: The participant responds to the latest message from a specific role. Useful for workflows where one role builds on another's output.
Invalid context base references (e.g., referencing a deleted role) automatically fall back to "Use Latest User Message".
Fan-out & Summarizer Patterns
The context base selector enables two powerful multi-agent patterns:
Fan-out Pattern
Configure multiple participants to use "Use Latest User Message". When you send a message:
- All participants respond to your input independently
- They don't see each other's outputs from the same turn
- Use case: Get multiple perspectives on the same question
Summarizer Pattern
Configure an early participant with "Use Latest User Message" (fan-out), and a later participant with "Follow Previous Message":
- The summarizer sees all responses from earlier participants
- Can synthesize or build upon their outputs
- Use case: Research team where specialists answer, then a coordinator synthesizes
Message Provenance
In the conversation history, each participant's response shows "Based on: [message]" indicating which message it's responding to. Tap the indicator to jump to that message.
Running a Flow
Once your flow is set up:
- Enable the flow by toggling "Use Custom Flow" in the Chat Flow tab
- Start a new conversation
- Type your message â all participants will respond in sequence
- Each participant sees the full conversation history, including other participants' responses
Chat Flow mode resets to default (off) when you start a new conversation, ensuring you don't accidentally use a flow when you want a simple chat.
Example Flows
Business Planning Team
- Market Analyst: Researches market trends and competition
- Strategic Planner: Develops strategy based on analysis
- Risk Assessor: Identifies potential risks and mitigation strategies
Content Creation Team
- Idea Generator: Brainstorms creative concepts
- Story Developer: Expands ideas into detailed content
- Editor: Reviews and polishes the final output
Code Review Team
- Code Reviewer: Reviews code for bugs and style issues
- Security Expert: Checks for security vulnerabilities
- Performance Analyst: Suggests optimization opportunities