AI - Chat Message
Type identifier:
ai:llm:chatMessageCategory: AI Operations
Description
The AI - Chat Message node enables conversations with large language models (LLMs) from multiple providers including OpenAI, Anthropic, Google AI, and Groq. It supports dynamic message templates with variable interpolation, streaming responses, extended thinking, and AI actions (tool use).
For exact supported model identifiers and provider-specific capability differences, use AI - Model Reference.
Input Handles
Handle | Type | Description |
|---|---|---|
| control | Triggers the LLM request. |
| string | (Dynamic mode) Runtime model name input. |
| varies | Dynamic handles for each defined variable. |
| number | Dynamic Temperature, when supported by the model. |
| number | Dynamic Top K, when supported by the model. |
| number | Dynamic Top P, when supported by the model. |
| number | Dynamic Max Tokens. |
| string[] | Dynamic Stop Sequences. |
| number | Dynamic thinking token budget. |
| any | Action result when Continue With Result is enabled. |
| control | Returns control from a continuing action. |
Output Handles
Handle | Type | Description |
|---|---|---|
|
| The complete response in non-streaming mode. |
| stream of strings | Text chunks in streaming mode, including when Parse to Content is enabled. |
|
| Thinking blocks when the configured model exposes thinking. |
| control | Runs the branch for a returned action call. |
| action Schema | Input returned for that action call. |
| control | Non-streaming response and action processing completed. |
| control | Request or response processing failed. |
Streaming mode exposes responseStream instead of responseText and success. Use Stream - Collect to process chunks and receive its done signal.
Configuration Options
Model Selection
Option | Type | Default | Description |
|---|---|---|---|
Model | enum or dynamic string handle |
| The LLM model to use, literal or runtime input. |
When Model is set to dynamic, the node exposes a model input handle and resolves the model name at runtime.
If the runtime model string is not in the supported model list, the node fails with a clear configuration error.
Available models are listed by provider (OpenAI, Anthropic, Google AI, and Groq) in AI - Model Reference — Chat Message Models. Deprecated model names and their replacements are in AI - Model Reference — Deprecated Models and Fallbacks.
Model Parameters
Parameter | Type | Description |
|---|---|---|
Temperature |
| Randomness in responses (0-2). |
Top K |
| Limit token selection to top K candidates. |
Top P |
| Nucleus sampling probability. |
Max Tokens |
| Maximum response length. |
Stop Sequences |
| Sequences that stop generation. |
Note: Sampling parameters (Temperature, Top K, Top P) are not configurable for adaptive-thinking Claude models (
claude-fable-5-1,claude-fable-5,claude-opus-5-5,claude-opus-5,claude-sonnet-5,claude-opus-4-8,claude-opus-4-7) and OpenAI non-chat/reasoning models (gpt-6-astra,gpt-6-sol,gpt-6-luna,gpt-5.6-sol,gpt-5.6-terra,gpt-5.6-luna,gpt-5.5,gpt-5.4,gpt-5.4-mini,gpt-5.4-nano,gpt-5.1,gpt-5-2025-08-07,gpt-5-mini-2025-08-07,gpt-5-nano-2025-08-07).
Response Options
Option | Type | Default | Description |
|---|---|---|---|
Parse to Content |
|
| Parse Markdown response as Content. |
Stream Response |
|
| Enable streaming response mode. |
Streaming responses buffer up to 64 unread chunks or 1,048,576 UTF-8 bytes, whichever limit is reached first. Reading pauses until chunks are consumed. A single chunk larger than the byte limit fails the stream. These limits do not cap the final accumulated response.
Extended Thinking
For supported models:
Option | Type | Description |
|---|---|---|
Enabled |
| Enable extended thinking mode. |
Tokens Budget |
| Maximum tokens for thinking. |
Adaptive thinking models such as claude-fable-5-1, claude-fable-5, claude-opus-5-5, claude-opus-5, claude-sonnet-5, claude-opus-4-8, and claude-opus-4-7 do not use the Tokens Budget option. Other supported Anthropic models use the configured Tokens Budget.
Auto-Continue
Option | Type | Description |
|---|---|---|
Max Turns |
| Maximum conversation turns for auto-continuation. |
Variables
Array of variables that can be used in message templates. Each variable creates a dynamic input handle.
Message Templates
Array of message templates defining the conversation:
Field | Type | Description |
|---|---|---|
Type |
| Message role. |
Message Template |
| Template with variable placeholders. |
Cached |
| Enable prompt caching (Anthropic). |
Repeating messages support:
- Iterate over array variables
- Create dynamic message sequences
Actions (Tool Use)
Field | Type | Description |
|---|---|---|
Name |
| Action name for the AI to call. |
Description |
| Description of what the action does. |
Input Schema | schema | Expected input structure. |
Force Use |
| Force the AI to use this action. |
Continue With Result |
| Auto-continue after action result. |
Behaviour
- Resolves all variable values from input handles
- Constructs messages from templates with interpolated variables
- Sends request to selected LLM provider
- Processes response (streaming or complete)
If actions are defined and invoked:
- Triggers corresponding action handle
- Waits for action result
- Optionally continues conversation with result
- In non-streaming mode, returns the final response and triggers
success
Variable Interpolation
Variables in templates are replaced with actual values:
- Content variables support rich formatting
- File variables are passed to vision-capable models
- Nested object properties accessed via dot notation
Provider-Specific Features
Feature | OpenAI | Anthropic | Groq | |
|---|---|---|---|---|
Streaming | ✓ | ✓ | ✓ | ✓ |
Vision | ✓ | ✓ | ✓ | ✓ |
Audio | ✓ | - | - | ✓ |
Thinking | - | ✓ | - | - |
Caching | - | ✓ | - | - |
Examples
Simple Question-Answer
Configuration:
- Model: claude-haiku-4-5-20251001
- Variables:
Question Messages:
- System: "You are a helpful assistant."
- User:
{Question}
Structured Extraction With Actions
Configuration:
- Model: gpt-4o
Actions:
- Name: "extract_data"
- Schema: Structured output format
- Force Use: true
- Continue With Result: false
Streaming Chat Response
Configuration:
- Model: claude-sonnet-5
- Stream Response: true
Use stream output with Stream - Collect to process incrementally.