Skip to main content

2. Creating Agents

This guide explains how to configure agents in Texterz.

Chatbot List

An Agent is the execution layer that connects knowledge, channels, and tools.

Channel Configuration​

Connect your agent to various channels like WhatsApp, Web Widget, or Telegram.

Bot Channels

Tool Integration​

Enable tools like Calendly or custom webhooks to extend your agent's capabilities.

Bot Tools Agents control behavior, goals, and permissions. They do not store knowledge themselves.

What an Agent Does​

An agent is responsible for:

  • Interpreting user messages
  • Querying assigned Knowledge Buckets
  • Triggering allowed tools
  • Returning responses through connected channels

Agents are channel-agnostic and reusable.

1. System Prompt (Behavior Definition)​

The System Prompt defines how the agent behaves.

Create Agent Step 1

Prompt Settings

It should clearly specify:

  • The agent’s role
  • The communication style
  • Explicit constraints

  • Role
    Describe the function the agent performs.
    Example:
    You are responsible for qualifying inbound leads for a solar installation company.

  • Tone
    Define how responses should be written.
    Example:
    Use professional, concise language. Format messages for chat interfaces.

  • Constraints
    Define what the agent must not do.
    Example:
    Do not provide information outside the assigned knowledge buckets.

Avoid vague instructions such as “be helpful” or “be smart”.

2. Defining the Agent Goal​

Each agent should have one primary objective.

Create Agent Step 2

Common goals include:

  • Lead qualification
  • Appointment scheduling
  • Customer support
  • Information routing to humans

Goals should be outcome-oriented and measurable
(e.g. “collect contact details” instead of “chat with users”).

3. Model Configuration​

Model Selection​

Choose a model based on task requirements:

Bot Settings

  • GPT-4o / Claude 3.5
    Use for complex reasoning, structured conversations, or accuracy-critical tasks.

  • Groq (Llama 3)
    Use when low latency is more important than deep reasoning.

Model choice affects cost, speed, and response quality.

Temperature​

Temperature controls response variability.

  • Low (0.1 – 0.3)
    Predictable, factual responses (support, FAQs).

  • Medium (0.5 – 0.7)
    Natural conversation (sales, qualification).

  • High (0.9+)
    Not recommended for production agents.

4. Assigning Knowledge Buckets​

Agents can only access explicitly assigned Knowledge Buckets.

Create Agent Step 3

Knowledge Settings

  • Select one or more buckets in the agent settings
  • The agent will not reference any other data
  • Knowledge access is enforced at runtime

This ensures clean separation between clients and use cases.

Advanced Agent Settings​

To learn more about Model Selection, Temperature, Conversation Memory, and Vision, visit the Advanced Bot Settings.

Before Moving On​

Verify that:

  • The system prompt is specific and constrained
  • The agent has a single clear goal
  • The correct knowledge buckets are attached
  • The selected model matches the task

Next Step​

Once the agent is created, connect it to a channel: