Search anything...
K
Back to Docs
  • Introduction
  • Quick Start
  • Account Setup
  • AI Studio
  • Chat
  • Agents
  • Voice
  • MCP Servers
  • Workflows
  • Authentication
  • Studio API
  • Chat API
  • Agents API
  • Voice API
  • Workflows API
  • Webhooks
  • Error Codes
  • Creating Custom Agents
  • MCP Integration
  • Building Workflows
  • Prompt Engineering
  • Team Management
  • Billing & Plans
  • Usage Monitoring
  • Single-Tenant Cloud
  • Private VPC Deployment
  • SSO Configuration
  • Security Policies
  • Compliance
  • Troubleshooting
  • API Versioning
DocsFeaturesAgents

Agents

Create and customize AI agents with specific personalities and capabilities

Agents are customizable AI assistants tailored to specific tasks and personalities. Create specialized agents for different use cases, configure their behavior, and integrate them with MCP tools.

Overview

Agents offer:

  • Custom Personalities - Define unique system prompts
  • Configurable Parameters - Adjust temperature and model
  • MCP Integration - Connect tools per agent
  • Persistent Configuration - Save and reuse agents
  • Dedicated Chat - Separate conversation per agent

Agent Concepts

What is an Agent?

An agent is a configured AI assistant with:

  • Identity - Name and description
  • Personality - System prompt defining behavior
  • Parameters - Model selection and temperature
  • Tools - Enabled MCP servers

System Prompts

The system prompt defines how your agent behaves:

You are a friendly customer support agent for a SaaS company.
Your primary goals are:
- Answer questions about our products
- Help troubleshoot common issues
- Escalate complex problems appropriately
- Maintain a helpful, professional tone

Always ask clarifying questions when needed and provide
step-by-step guidance when explaining solutions.

Creating an Agent

Step 1: Basic Information

  1. Navigate to Agents in the sidebar
  2. Click New Agent
  3. Fill in basic details:
FieldDescriptionExample
NameAgent's display name"Marketing Writer"
DescriptionBrief agent description"Creates marketing copy and content"
AvatarVisual identifier (emoji)"✍️"

Step 2: System Prompt

Write a detailed system prompt that defines your agent:

You are an expert marketing copywriter with 15 years of experience.

## Your Expertise
- Compelling headlines and hooks
- Product descriptions that sell
- Email marketing campaigns
- Social media content
- Landing page copy

## Your Style
- Clear and concise
- Benefit-focused
- Action-oriented
- Brand-appropriate

## Guidelines
- Ask about the target audience before writing
- Suggest A/B testing variations
- Include calls-to-action
- Consider SEO when relevant

Step 3: Model Configuration

SettingOptionsDescription
ModelGemini 2.0 Flash, Gemini 1.5 ProAI model to use
Temperature0.0 - 1.0Controls creativity

Temperature Guide:

ValueBehaviorBest For
0.0 - 0.3Focused, deterministicCode, factual content
0.4 - 0.6BalancedGeneral assistance
0.7 - 1.0Creative, variedCreative writing, brainstorming

Step 4: MCP Tools

Enable MCP servers for your agent:

  1. View available servers in the configuration
  2. Toggle servers on/off
  3. Each agent can have different tools enabled

Tool Selection Tips:

  • Enable only relevant tools
  • Consider the agent's purpose
  • More tools = more capabilities but potential complexity

Step 5: Save Agent

  1. Review all settings
  2. Click Save Agent
  3. Agent appears in your agent list

Using Agents

Starting a Chat

  1. Select an agent from the grid
  2. Click or tap to open the chat interface
  3. Begin conversing with your agent

Agent Chat Interface

┌─────────────────────────────────────────────────────────┐
│  ← Back to Agents          Marketing Writer ✍️          │
├─────────────────────────────────────────────────────────┤
│                                                         │
│  ┌─────────────────────────────────────────────────┐   │
│  │  You: I need help writing a product launch      │   │
│  │  email for our new project management tool.     │   │
│  └─────────────────────────────────────────────────┘   │
│                                                         │
│  ┌─────────────────────────────────────────────────┐   │
│  │  Agent: I'd be happy to help! Before I start,  │   │
│  │  could you tell me:                             │   │
│  │  1. Who is your target audience?                │   │
│  │  2. What's the key differentiator?              │   │
│  │  3. Is there a special launch offer?            │   │
│  └─────────────────────────────────────────────────┘   │
│                                                         │
│  ─────────────────────────────────────────────────────  │
│  [ Type your message...                        ] [Send] │
└─────────────────────────────────────────────────────────┘

Conversation Management

Each agent maintains its own conversation:

  • Context Preserved - Previous messages remembered
  • Independent History - Separate from other agents
  • Persistent - Survives page refresh

Agent Templates

Customer Support Agent

You are a friendly and efficient customer support representative.

## Core Responsibilities
- Answer product questions accurately
- Troubleshoot issues step-by-step
- Process requests professionally
- Escalate when necessary

## Communication Style
- Empathetic and patient
- Clear and concise
- Solution-focused
- Professional yet warm

## Guidelines
- Always greet the customer
- Confirm understanding before solving
- Provide clear next steps
- Offer additional help

Settings: Temperature 0.3, Model: Gemini 2.0 Flash

Code Review Agent

You are an expert code reviewer with deep knowledge of
software engineering best practices.

## Review Focus Areas
- Code quality and readability
- Performance implications
- Security vulnerabilities
- Best practices adherence
- Test coverage

## Review Style
- Constructive and educational
- Specific, actionable feedback
- Explain the "why" behind suggestions
- Acknowledge good patterns

## Guidelines
- Prioritize issues by severity
- Suggest improvements, don't just criticize
- Reference documentation when helpful
- Consider context and constraints

Settings: Temperature 0.2, Enable GitHub MCP

Creative Writing Agent

You are a creative writing coach and collaborator with
expertise in various genres and styles.

## Capabilities
- Story development and plotting
- Character creation
- Dialogue writing
- World-building
- Style and voice guidance

## Approach
- Encourage experimentation
- Provide constructive feedback
- Offer alternatives, not prescriptions
- Inspire creativity

## Style
- Enthusiastic and supportive
- Rich vocabulary
- Engaging and imaginative

Settings: Temperature 0.9, Model: Gemini 1.5 Pro

Data Analyst Agent

You are a data analysis expert who helps interpret data
and generate insights.

## Expertise
- Statistical analysis
- Data visualization recommendations
- SQL query assistance
- Insight generation
- Report writing

## Approach
- Ask clarifying questions about data
- Explain findings clearly
- Provide actionable insights
- Suggest visualizations

## Guidelines
- Verify data understanding first
- Use appropriate statistical methods
- Explain confidence levels
- Note limitations and assumptions

Settings: Temperature 0.3, Enable PostgreSQL MCP

Research Assistant

You are a thorough research assistant who helps gather
and synthesize information.

## Capabilities
- Information gathering
- Source evaluation
- Summary creation
- Citation tracking
- Comparative analysis

## Research Process
1. Clarify research questions
2. Identify relevant sources
3. Extract key information
4. Synthesize findings
5. Present conclusions

## Guidelines
- Cite sources when possible
- Note information gaps
- Present balanced viewpoints
- Highlight key findings

Settings: Temperature 0.4, Enable Web Search MCP

Managing Agents

Editing Agents

  1. Click the Edit button on an agent card
  2. Modify settings as needed
  3. Click Save Changes

Duplicating Agents

Create variations of existing agents:

  1. Open the agent you want to copy
  2. Click Duplicate
  3. Modify the copy as needed
  4. Save with a new name

Deleting Agents

  1. Click the Delete button on an agent card
  2. Confirm deletion
  3. Agent and its chat history are removed

Best Practices

Writing Effective System Prompts

  1. Be Specific - Detail exact behaviors
  2. Use Structure - Organize with headers/lists
  3. Set Boundaries - Define what agent should/shouldn't do
  4. Include Examples - Show desired responses
  5. Consider Edge Cases - Handle unusual requests

Choosing Temperature

Use CaseRecommended Temperature
Customer support0.2 - 0.4
Technical writing0.2 - 0.3
General assistance0.5 - 0.6
Brainstorming0.7 - 0.8
Creative writing0.8 - 1.0

Tool Integration

  • Match tools to purpose - Code agent needs GitHub, not Slack
  • Start minimal - Add tools as needed
  • Test combinations - Verify tools work well together
  • Document tool usage - Include in system prompt

Troubleshooting

Agent Not Following Instructions

  • Make system prompt more explicit
  • Lower temperature for consistency
  • Add specific examples
  • Reinforce guidelines in conversation

Slow Responses

  • Use Gemini 2.0 Flash for speed
  • Reduce enabled MCP tools
  • Simplify system prompt

Inconsistent Behavior

  • Lower temperature value
  • Add more specific guidelines
  • Use structured system prompts
  • Test with varied inputs

Previous: Chat | Next: Voice

Previous
Chat
Next
Voice