---
title: LLM Chat
description: Run text through a language model for writing, reasoning, extraction, or transformation.
source: https://www.edgaze.ai/docs/builder/nodes/llm-chat
section: builder
---
# LLM Chat

> Run text through a language model for writing, reasoning, extraction, or transformation.

## Overview

LLM Chat is the main text-generation block. It accepts upstream data plus instructions, runs a selected model, and returns a response for the next step.

Use it for rewriting, summarising, classifying, analysing, or any job that needs language understanding. Default model is Claude Sonnet 4.6 for strong creator quality.

```docsgraph
llm-chat
```

## Ports

- **Input , Input:** The data to work on , connected upstream text or JSON-like content.
- **Output , Response:** The model's generated text for downstream blocks.

## Inspector

- **Prompt:** Instructions used when Input is not connected, or as guidance alongside connected data.
- **Style / System (optional):** Tone, role, and constraints for the model.
- **Model:** Choose among GPT, Claude, or Gemini options , trade off quality, cost, and speed. Match the provider key in the run modal.
- **Temperature:** Lower for precise extraction; higher for creative variation.
- **Max Tokens:** Cap on response length , align with what the product should return.
- **Stream** and **Safe Mode:** Inline toggles for live streaming and safer defaults.

## Tips

- Give each LLM block one clear job instead of stacking many tasks in one prompt.
- Feed raw data through Input; keep transformation instructions in the prompt.
- Use a stronger model for nuanced writing; lighter models for fast utility work.
- Set max tokens close to the expected answer length to avoid waste.

## Related

- [Workflow Studio](/docs/builder/workflow-studio)