Puzzlet supports extending prompts with custom tools (like API calls or calculations) and creating agents that can make multiple LLM calls to solve complex tasks.

​
Tools

Tools are custom functions that your prompts can use to interact with external systems or perform specific operations.

​
1. Create and Register Tools

First, create and register your tools with the SDK:

import { ToolPluginRegistry } from "@puzzlet/agentmark";

// Create your tool function
const checkWeather = async (input: { location: string, date?: string }) => {
  // Your weather checking logic here
  const weather = await weatherAPI.get(input.location, input.date);
  return { temperature: weather.temp, conditions: weather.conditions };
};

// Register the tool
ToolPluginRegistry.register(checkWeather, "check_weather");

​
2. Define Tool Schema in Prompt

Then use the tool in your prompt:

---
name: weather-advisor
metadata:
  model:
    name: gpt-4
    settings:
      tools:
        check_weather:
          description: Get current weather for a location
          parameters:
            type: object
            properties:
              location:
                type: string
                description: City or location name
              date:
                type: string
                description: Date to check (YYYY-MM-DD)
            required: ["location"]
---

<System>
You are a travel advisor that can check weather conditions. Use the weather tool to provide accurate forecasts.
</System>

<User>What's the weather like in Paris today?</User>

​
Agents

Agents can make multiple LLM calls and use tools to solve complex tasks.

​
1. Register Multiple Tools

import { ToolPluginRegistry } from "@puzzlet/agentmark";

// Register all your tools
ToolPluginRegistry.register(searchFlights, "search_flights");
ToolPluginRegistry.register(checkWeather, "check_weather");

​
2. Create Multi-Step Agent

Enable agent capabilities by setting max_llm_calls in your prompt:

---
name: travel-planner
metadata:
  model:
    name: gpt-4
    settings:
      max_llm_calls: 3  # Allow multiple LLM calls
      tools:
        search_flights:
          description: Search available flights
          parameters:
            type: object
            properties:
              from:
                type: string
              to:
                type: string
              date:
                type: string
            required: ["from", "to"]
        check_weather:
          description: Check weather forecast
          parameters:
            type: object
            properties:
              location:
                type: string
            required: ["location"]
---

<System>
You are a travel planner that can:
1. Search for flights
2. Check destination weather
3. Make travel recommendations
</System>

<User>Help me plan a trip from NYC to Miami next week.</User>

​
3. Run the Agent

import { runInference } from '@puzzlet/agentmark';

const result = await runInference(prompt, {
  // Your props here
});

​
Learn More

For detailed information about implementing tools, creating agents, and best practices, see the AgentMark Tools and Agents Documentation.

Key features include:

  • Custom tool creation
  • Multiple tool support
  • Multi-step agent workflows
  • Error handling
  • Parameter validation

Have Questions?

We’re here to help! Choose the best way to reach us:

Was this page helpful?

JSON OutputOverview