Saltar al contenido principal

Custom Agents

Learn how to create fully custom AI agents with complete control over their behavior, personality, and capabilities.

What are Custom Agents?​

While AgentsGT provides pre-built templates (Marketing Consultant, Generalist Assistant, Pre-sales Consultant, Tech Support), custom agents let you build an agent from scratch with your own system prompt, personality, and tool configuration.

When to use custom agents:

  • Your use case doesn't match any template
  • You need precise control over the agent's behavior
  • You want to combine capabilities from multiple templates
  • You need specialized tool configurations

Creating a Custom Agent​

Step 1: Start Agent Creation​

  1. Navigate to your organization's agents page
  2. Click "Create Agent"
  3. Toggle "Custom Prompt" mode instead of selecting a template

Step 2: Write Your System Prompt​

The system prompt is the foundation of your custom agent. It defines everything about how the agent behaves.

A good system prompt includes:

You are [specific role] for [company/context].

Your responsibilities:
- [What the agent should do]
- [What information it should provide]
- [How it should handle common scenarios]

Guidelines:
- [Tone and personality rules]
- [Boundaries and limitations]
- [Response format preferences]

You must NOT:
- [Things the agent should never do]
- [Topics to avoid]
- [Actions that require human escalation]

Example — E-commerce Support Agent:

You are a customer support specialist for TechGadgets, an online electronics store.

Your responsibilities:
- Answer questions about product features, specifications, and compatibility
- Help customers track orders using their order ID
- Process return and exchange requests
- Recommend products based on customer needs and budget

Guidelines:
- Be friendly and professional at all times
- Keep responses concise — under 3 paragraphs unless detailed instructions are needed
- Always ask for the order ID when handling order-related questions
- Suggest related products when appropriate, but don't be pushy

You must NOT:
- Process refunds directly — escalate to human support
- Share other customers' information
- Make promises about future product releases
- Provide medical, legal, or financial advice

Step 3: Configure Settings​

Model Selection​

Choose the AI model based on your needs:

  • Fast models — Best for quick, simple interactions (FAQ bots, basic support)
  • Creative models — Best for content generation, marketing copy, brainstorming
  • Precise models — Best for technical support, data analysis, accuracy-critical tasks

Language​

Set the primary language for the agent. The agent can still understand and respond in other languages if asked, but will default to the selected language.

Tone​

Select a conversation tone:

  • Professional — Formal business communication
  • Friendly — Warm and approachable
  • Casual — Relaxed and conversational
  • Concise — Brief, to-the-point responses

Greeting Message​

Set the first message users see when starting a conversation:

Hi! I'm your TechGadgets support assistant. I can help you with:
- Product questions and recommendations
- Order tracking
- Returns and exchanges

How can I help you today?

Conversation Starters​

Add suggested prompts that users can click to start a conversation:

  • "Track my order"
  • "Help me choose a laptop"
  • "Start a return"

Step 4: Add Knowledge (Optional)​

Upload files to give your agent specialized knowledge:

  • Product catalogs (PDF, CSV)
  • Company policies (DOCX, TXT)
  • Technical documentation
  • FAQs and help articles

See Creating Your First Agent for detailed instructions on adding files.

Step 5: Configure Tools​

Enable or disable specific tools for your agent:

  • Web Search — Allow the agent to search the web for current information
  • Image Generation — Let the agent create images
  • RAG (File Search) — Enable the agent to search uploaded documents
  • Google Drive — Access files from connected Google Drive

Toggle only the tools your agent needs to keep responses focused.

Step 6: Advanced Settings​

Custom Slug​

Create a URL-friendly identifier for your agent. This is used in API endpoints and embed URLs:

  • Use lowercase letters and hyphens
  • Keep it short and descriptive
  • Example: tech-support, sales-bot, onboarding-helper

Additional Instructions​

Add supplementary instructions that extend the system prompt. Use this for:

  • Seasonal updates ("We're running a 20% off sale this month")
  • Temporary changes ("Our New York office is closed for renovation")
  • Context that changes frequently

Testing Your Custom Agent​

After creating your agent:

  1. Go to the agent's detail page
  2. Use the built-in chat interface to test
  3. Try various scenarios:
    • Common questions your users would ask
    • Edge cases and unusual requests
    • Questions outside the agent's scope
    • Multi-turn conversations

Iterate based on results:

  • If the agent is too verbose → Add "Keep responses concise" to your system prompt
  • If the agent hallucinates → Add "If you don't know, say so" and upload more context files
  • If the agent goes off-topic → Add clear boundaries in the "You must NOT" section
  • If the tone is wrong → Adjust the tone setting or add tone instructions to the prompt

Best Practices​

  1. Start specific, then broaden — Begin with a narrow scope and expand as you test
  2. Use examples in your prompt — Show the agent how you want it to respond
  3. Set clear boundaries — Define what the agent should NOT do
  4. Keep context relevant — Only upload files the agent needs
  5. Test with real scenarios — Use actual customer questions for testing
  6. Iterate frequently — Refine the prompt based on real conversations

What's Next?​