Prompt

What is a Prompt?

A prompt is the structured input sent to an LLM to instruct it on what to do. In the context of Axiomkit, prompts go beyond basic user queries - they include full context, memory, tool access, and expected response formats.

Simple Prompts vs. Structured Agent Prompts

❌ Basic Prompt (Like ChatGPT)

User: I’m having trouble with my order.
Assistant: Sorry to hear that. Can you give me more details?

This response is passive - no action is taken, no context is stored, and the system doesn’t move forward.

✅ Axiomkit Agent Prompt (Structured for Action)

You are a support agent with access to:
- Create support tickets
- Send chat replies
- Email customers
- Recall user history

Current state:
- User asked: "My order hasn't arrived."
- Available actions: createTicket, sendChat, sendEmail
- Context: user321 in live chat

Respond with:
<action_call name="createTicket">{"issue": "Delayed order", "userId": "user321"}</action_call>
<output type="chat:send">I’ve created a support ticket for your delayed order. We’ll follow up via email.</output>

Why Structured Prompts Matter

Unstructured LLMs:

Don’t know what tools (actions) are available
Forget prior user preferences or history
Can’t reliably follow multi-step instructions
Only generate text - they don’t execute behavior

Structured prompts solve this by giving the LLM clear, machine-readable guidance on what it can do, what context it's in, and how to respond.

Every time your agent thinks, Axiomkit automatically builds a prompt like this:

1. Instructions

You are a customer support agent. Analyze the incoming message, decide what action to take, and respond appropriately.

2. Available Tools

<available-actions>
  <action name="createTicket">
    <description>Creates a new support ticket for a reported issue</description>
    <schema>{"type":"object","properties":{"issue":{"type":"string"},"userId":{"type":"string"}}}</schema>
  </action>
</available-actions>

<available-outputs>
  <output type="chat:send">
    <description>Sends a real-time message to the user</description>
    <schema>{"type":"string"}</schema>
  </output>
</available-outputs>

3. Context State

<contexts>
  <context type="chat" key="user321">
    user321: "My order hasn't arrived."
    agent: "I’m sorry to hear that—checking now."
  </context>
</contexts>

4. Updates

<updates>
  <input type="chat:message" timestamp="2025-07-27T10:30:00Z">
    My order hasn’t arrived.
  </input>
</updates>

5. Expected Response Format

<response>
  <reasoning>Your internal logic</reasoning>
  <action_call name="createTicket">{"issue": "Delayed order", "userId": "user321"}</action_call>
  <output type="chat:send">Response to the user</output>
</response>

LLM Response Example

<response>
  <reasoning>
    The user reports a delayed order. I should create a support ticket and notify them in chat.
  </reasoning>
  <action_call name="createTicket">{"issue": "Delayed order", "userId": "user321"}</action_call>
  <output type="chat:send">I’ve opened a support ticket. You’ll receive updates by email.</output>
</response>

AxiomKit automatically:

Executes the createTicket handler
Sends the chat message using the sendChat output
Logs the LLM’s reasoning for traceability

Benefits of Axiomkit Prompting

✅ Actionable Intelligence: The LLM doesn’t just reply-it decides and acts.
✅ Consistency: All agents follow a shared, structured schema.
✅ Contextual Awareness: Prompts include relevant memory and history.
✅ Debuggable: Every reasoning chain and decision is traceable.
✅ Flexible: You can easily plug in new actions, outputs, and context types.

The structured prompting system is what transforms an LLM from a passive text generator into an intelligent, autonomous agent capable of taking actions.

Prompt

On this page