AI Agent

The AI Agent is the heart of InfuseOS. It understands natural language, executes multi-step tasks, and shows you exactly what it's doing in real-time.

How It Works

When you send a message, the AI Agent:

1. Understands your intent - Parses your natural language request
2. Plans the approach - Determines what steps are needed
3. Executes tools - Runs integrations, searches, and other actions
4. Synthesizes results - Combines outputs into a coherent response

You can watch this process in real-time as the agent shows each step it's taking.

Available AI Models

InfuseOS offers multiple AI models optimized for different tasks:

Choosing the Right Model

• GPT 5.1: Great for everyday tasks. Fast and efficient. Included in Free plan.
• GPT 5.2: Better at complex reasoning and multi-step analysis.
• Gemini 3 Pro: Excellent for tasks requiring long context windows.
• Gemini 3 Flash: Fastest responses for simple queries.

Reasoning Levels

For complex problems, you can adjust how deeply the AI thinks before responding:

How Reasoning Works

Different models implement reasoning differently:

• OpenAI models: Use "reasoning effort" parameter
• Google models: Use "thinking config" with budget levels
• Anthropic models: Use token budgets for extended thinking

The higher the reasoning level, the more thorough the analysis—but responses take longer.

Tool Execution

The AI Agent can execute various tools to accomplish tasks:

Core Tools

• Web Search: Find current information online
• Web Fetch: Retrieve content from specific URLs
• File Analysis: Process uploaded documents and images

Integration Tools

• Gmail: Send, search, and read emails
• Calendar: Manage events and find free time
• Drive: Access and manage files
• Docs: Create and edit documents

Knowledge Tools

• Search Memory: Find saved information
• Save to Memory: Store important content

Viewing Tool Execution

As the agent works, you'll see tool execution cards showing:

• Tool name: What action is being performed
• Status: Running, completed, or failed
• Results: Summary of what was found or done

Approval System

For security, some actions require your approval before execution:

Actions Requiring Approval

How Approval Works

1. The agent proposes an action
2. You see a preview of what will happen
3. Click Approve to proceed or Deny to cancel
4. The action executes only after approval

Real-Time Activity Timeline

Watch the agent work in real-time:

• Thinking indicator: Shows when the AI is processing
• Tool cards: Display as each tool is called
• Progress updates: See results as they come in
• Final response: Synthesized answer after all steps complete

Multi-Step Task Execution

The agent excels at complex, multi-step tasks:

Example: "Find a free slot next week, schedule a meeting with John, and email him the details"

The agent will:

1. Search your calendar for free slots
2. Create a new event at an available time
3. Compose and send an email to John
4. Report back with confirmation

Each step is visible in the conversation, so you always know what's happening.

Tips for Best Results

1. Be Specific

❌ "Check my emails"
✅ "Search my inbox for emails from John about the project proposal from last week"

2. Provide Context

❌ "Schedule a meeting"
✅ "Schedule a 30-minute meeting with the marketing team for Thursday afternoon to discuss Q4 planning"

3. Combine Related Tasks

❌ Three separate messages
✅ "Check my calendar for tomorrow, find any conflicts, and email me a summary"

4. Use Higher Reasoning for Complex Tasks

For important decisions or complex analysis, increase the reasoning level to "High".

Credit Usage

AI Agent interactions cost credits:

Learn more about credits →

Related