> ## Documentation Index
> Fetch the complete documentation index at: https://www.adaline.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.
# Use tools in prompts
> Enable function calling and tool use to extend your LLM's capabilities with external services
Tools let your LLM interact with external services, databases, and APIs during a conversation. When a model determines it needs external data or actions, it generates a structured tool call request that can be executed to fetch results and continue the conversation.
Adaline supports the full tool calling workflow — from defining tool schemas to configuring automatic execution with custom backends.
## Add tools to a prompt
Choose an LLM that supports tool calling (function calling). Most modern models from OpenAI, Anthropic, and Google support this feature.
Compose a prompt that may require external data or actions. For example, a prompt asking about current weather conditions would benefit from a weather API tool.
Enable the tool choice feature in the model settings to allow the LLM to generate tool calls.
Set the tool choice mode to control how the model uses your tools.
Choose from the following modes (availability varies by model):
| Mode | Behavior |
| ------------ | --------------------------------------------------------------------------------- |
| **none** | The model will not invoke any tools. |
| **auto** | The model decides which tools to use and when, based on the conversation context. |
| **required** | The model must invoke at least one tool in its response. |
| **any** | The model can call any of the available tools. |
Click **Add Tool** to create a new inline tool definition, or link to an already defined tool in the project.
## Tool schema definition
Each tool is defined using a JSON schema that tells the model what the tool does and what parameters it accepts. Click **Add Tool** to open the schema editor:
Here is the complete JSON structure for a tool definition:
```json theme={null}
{
"type": "function",
"definition": {
"schema": {
"name": "get_weather",
"description": "Get the current weather in a given location",
"parameters": {
"type": "object",
"properties": {
"location": {
"type": "string",
"description": "The city and state, e.g. San Francisco, CA"
},
"unit": {
"type": "string",
"enum": ["celsius", "fahrenheit"]
}
},
"required": ["location"]
}
}
}
}
```
### Schema reference
| Field | Type | Required | Description |
| ----------------------- | ------ | -------- | -------------------------------------------------------------------- |
| `type` | String | Yes | Always set to `"function"`. |
| `name` | String | Yes | A unique identifier for the tool. |
| `description` | String | Yes | Describes what the tool does. Helps the model decide when to use it. |
| `parameters.type` | String | No | The parameter structure type. Typically `"object"`. |
| `parameters.properties` | Object | No | Defines individual parameters with their types and descriptions. |
| `required` | Array | No | Lists the parameter names that are mandatory. |
Refer to the [OpenAI function calling documentation](https://platform.openai.com/docs/guides/function-calling) for more schema examples and best practices.
## Add tool calls and responses to messages
Beyond defining tools, you can add tool call and tool response messages directly in the Editor. This is useful for building multi-shot prompts that demonstrate how the model should interact with tools.
A **tool call** (in an Assistant message) represents the model invoking a specific tool. A **tool response** (in a Tool message) shows the data returned by the external service. Together, they teach the model the expected interaction pattern.
## Configure auto tool calls
For tools that connect to a live backend, you can configure an HTTP request endpoint so the [Playground](/docs/iterate/tool-calls-in-playground) automatically executes the tool and continues the conversation.
Add a `request` object to your tool definition:
```json theme={null}
{
"type": "function",
"definition": {
"schema": {
"name": "get_weather",
"description": "Get the current weather in a given location",
"parameters": {
"type": "object",
"properties": {
"location": {
"type": "string",
"description": "The city and state, e.g. San Francisco, CA"
}
},
"required": ["location"]
}
}
},
"request": {
"type": "http",
"method": "get",
"url": "https://api.my-tool-backend.com/get_weather",
"headers": {
"Content-Type": "application/json",
"Authorization": "Bearer "
},
"retry": {
"maxAttempts": 3,
"initialDelay": 1000,
"exponentialFactor": 2
}
}
}
```
### Request configuration reference
| Field | Type | Required | Description |
| ------------------------- | ------ | -------- | ----------------------------------------------------- |
| `type` | String | Yes | Always set to `"http"`. |
| `method` | String | Yes | The HTTP method (`GET` or `POST`). |
| `url` | String | Yes | The endpoint URL to call. |
| `headers` | Object | No | HTTP headers to include in the request. |
| `retry` | Object | No | Retry configuration for failed requests. |
| `retry.maxAttempts` | Number | No | Maximum number of retry attempts. |
| `retry.initialDelay` | Number | No | Initial delay in milliseconds before the first retry. |
| `retry.exponentialFactor` | Number | No | Multiplier for exponential backoff between retries. |
When auto tool calls are enabled and a tool has a configured request endpoint, the Playground will automatically invoke the tool, inject the response, and continue the conversation — enabling fully automated multi-turn interactions.
## Next steps
Test tool interactions in the Playground sandbox.
Connect to MCP servers for standardized tool access.