Chat

Natural conversations with AI agents for interactive task execution and collaboration.

What is Chat?

Chat provides real-time interaction with AI agents to accomplish tasks through natural conversation. Unlike workflows (automated sequences) or tasks (scheduled operations), chat enables dynamic, conversational task execution with immediate feedback.

Key Capabilities:

• Multi-agent collaboration in single conversation
• File and attachment support
• Real-time tool execution visibility
• Context-aware conversations
• Shareable chat sessions

Multi-Agent Communication

Agent Selection: Chat supports multiple agents working together in a single conversation.

How It Works:

1. Click agent selector dropdown in chat input
2. Select one or multiple agents
3. All selected agents receive your messages
4. Agents collaborate based on their specialties
5. Remove agents by clicking X on their pill

Default Behavior:

• All published agents selected by default
• AI routes messages to most relevant agent(s)
• Agents access shared conversation context

Agent Pills: Selected agents appear as removable pills above input field.

Use Cases:

• Research + Writing: Research agent gathers data, writing agent creates content
• Code + Review: Development agent writes code, reviewer validates
• Multi-language: Different agents for different languages
• Specialized Teams: Sales, support, and technical agents collaborating

Starting a Chat

Create New Chat:

1. Navigate to Chat page
2. Type message in input field
3. Press Enter or click Send
4. Chat session auto-created with unique ID

Chat Components:

• Chat History: Left sidebar (collapsible)
• Message Area: Center conversation view
• Input Field: Bottom message composer
• Tool Panel: Right sidebar for tool execution details
• Settings: Model configuration and preferences

Sending Messages

Text Messages:

• Type in input field
• Markdown supported
• Press Enter to send (Shift+Enter for new line)

Attachments:

• Click attachment icon
• Supports: images, documents, code files
• Multiple attachments per message
• Files visible in message history

Voice Input: Voice input capabilities for audio-based interaction.

Agent Responses

Response Types:

• Text: Direct answers and explanations
• Tool Calls: External API calls, database queries, calculations
• Code Execution: Running JavaScript/Python
• File Generation: Creating documents, images, data files

Tool Execution: When agents use tools:

1. Tool call appears in message
2. Click to view in tool panel (right sidebar)
3. Real-time execution status
4. Results incorporated into response

Response Streaming: Messages stream in real-time as agents generate responses.

Chat Settings

Model Selection:

• Choose AI model (GPT-4, Claude, Gemini, etc.)
• Different capabilities per model
• Workspace default or custom selection

Configuration:

• Temperature: Creativity level (0-1)
• Max Tokens: Response length limit
• Instructions: Custom system prompt
• Reasoning: Enable chain-of-thought

Task Planning: Toggle task planning for structured multi-step responses.

Chat History

Features:

• Chronological list of chat sessions
• Search by content
• Quick access to recent chats
• Auto-save all conversations

Navigation: Click any chat to load conversation. New messages append to active chat.

Tool Execution Panel

Real-time Tool Visibility: See exactly what agents are doing:

• API calls being made
• Code being executed
• Data being processed
• External services accessed

Tool Types:

• HTTP Requests: API calls with request/response
• Code Execution: JavaScript/Python runtime
• Database Queries: Data retrieval
• File Operations: Read/write files
• Composio Actions: Third-party integrations

Expanding Panel: Click tool calls to view detailed execution information.

File Browser

When Available: For chats with sandbox/workspace access:

• Browse workspace files
• View file contents
• Reference files in conversation
• Agent can read/modify files

Access: Click folder icon in chat header.

Best Practices

Effective Communication:

• Be specific about desired outcomes
• Provide context for complex tasks
• Use attachments for reference materials
• Break complex tasks into steps

Agent Selection:

• Use all agents for broad tasks requiring diverse skills
• Select specific agents for specialized tasks
• Remove irrelevant agents to reduce noise
• Test different agent combinations

Tool Usage:

• Monitor tool panel to understand agent actions
• Review tool outputs for accuracy
• Trust but verify external API calls

Conversation Management:

• Start new chat for unrelated topics
• Use descriptive first messages (becomes chat title)
• Share valuable conversations with team
• Archive completed conversations

Use Cases

Quick Tasks: Single-turn requests like “summarize this document” or “generate code snippet”.

Interactive Development: Back-and-forth refinement of code, content, or analysis.

Research & Analysis: Multi-step investigation with agent collaboration.

Learning & Exploration: Ask questions, explore concepts, get explanations.

Team Collaboration: Shared problem-solving with shareable conversations.

Next Steps

• Explore AI Agents to create specialized agents
• Review Workflows for automated processes
• Check Tasks for scheduled operations