> ## Documentation Index
> Fetch the complete documentation index at: https://docs.workmore.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# AI

> Steps that use a language model to read, write, classify, or decide.

Steps that use a language model to read, write, classify, or decide.

## AI / LLM

<code>ai.llm</code> · action

Call an AI language model to summarize, extract, generate, classify, or analyze data. Use presets for common tasks or provide a custom system prompt.

**Settings**

| Setting       | Required | What it does                                                                                                                                        |
| ------------- | -------- | --------------------------------------------------------------------------------------------------------------------------------------------------- |
| Provider      | No       | Select which LLM provider to use. The credential is resolved from the matching App Management connection (per-canvas default or per-node override). |
| Model Tier    | No       | Choose the model tier based on task complexity. Fast is good for simple tasks, Powerful for complex reasoning.                                      |
| Task Preset   | No       | Select a preset to use an optimized system prompt for common tasks, or choose Custom to write your own.                                             |
| System Prompt | No       | Custom instructions for the AI. Only used when Task Preset is set to Custom.                                                                        |
| User Message  | Yes      | The main prompt or question for the AI. Reference upstream data using template syntax.                                                              |
| Context Data  | No       | Data from upstream nodes to include in the prompt. Will be serialized and appended to the user message.                                             |

**Produces**

* `response` — Response
* `tokens_used` — Tokens Used
* `model` — Model
* `duration_ms` — Duration (ms)
