ComfyUI LLM Assistant

AI assistant sidebar for ComfyUI.

Creator: CrazyDashTool

Version: 1.2.1

Features

• Multi-chat history: create, switch, delete, and keep separate chat sessions.
• Context-aware answers: the assistant can include the current workflow, saved workflows, installed models, and recent chat history.
• Streaming responses over SSE: text appears while the model is generating.
• Cancel button for long generations.
• Markdown rendering for assistant replies: headings, lists, inline code, code blocks, copy buttons, and workflow JSON apply buttons.
• Prompt Enhancer tab: expand a simple prompt into a detailed ComfyUI prompt with weights, style, camera, lighting, negative prompt, and technical parameters.
• Insert enhanced prompts into the selected ComfyUI node when it has a compatible text/prompt widget.
• Image upload for vision-capable models/providers.
• Ollama Web Search context: optional web search results are injected into the prompt before the LLM call.
• Voice input through the browser Web Speech API.
• ComfyUI-style dark interface instead of the older futuristic look.
• File/model browser and beginner course builder are still included.

Supported Providers

Provider	Key	Default model
OpenAI	Required	gpt-5.5
Anthropic Claude	Required	claude-sonnet-4-6
Google Gemini	Required	gemini-3-pro-preview
Groq	Required	openai/gpt-oss-120b
xAI Grok	Required	grok-4.20-reasoning
OpenRouter	Required	minimax/minimax-m2.5:free
Mistral API	Required	mistral-medium-3-5
Ollama local	Not required	local-models
LM Studio local	Not required	local-models

Model fields are editable. Some model names are aliases or preview models and may depend on your provider account access.

Ollama Web Search

Ollama Web Search uses Ollama's hosted endpoint:

POST https://ollama.com/api/web_search
Authorization: Bearer <OLLAMA_API_KEY>

This is separate from the local Ollama server. Create a free Ollama account key, put it into Settings -> Ollama Web Search API key, and enable Ollama web search below the chat input.

Installation

1. Download or clone the repository.
2. Place the folder into ComfyUI/custom_nodes/.

ComfyUI/
  custom_nodes/
    ComfyUI-LLM-Assistant/
      __init__.py
      server.py
      web/
        llm_assistant.js
      requirements.txt
      README.md

3. Install dependencies if needed:

pip install -r requirements.txt

4. Restart ComfyUI.
5. Open the LLM Assistant sidebar tab or the floating AI button.

Local Models

Ollama

1. Install Ollama.
2. Pull a model, for example:

ollama pull qwen3:8b

3. Select Ollama local in the assistant settings.
4. Default Base URL: http://127.0.0.1:11434.
5. Use a vision-capable local model if you want image input.

LM Studio

1. Open LM Studio.
2. Download and load a model.
3. Turn on the Local Server.
4. Select LM Studio local in the assistant settings.
5. Default Base URL: http://127.0.0.1:1234.

API Endpoints

Method	Path	Description
POST	/llm-assistant/chat	Non-streaming chat fallback
POST	/llm-assistant/chat/stream	SSE streaming chat
POST	/llm-assistant/web-search	Ollama web search helper
GET	/llm-assistant/files	ComfyUI folder structure
GET	/llm-assistant/models-list	Installed model files
GET	/llm-assistant/workflow-files	Saved workflow JSON files

Security

• Provider keys are stored only in browser localStorage.
• Requests go through the local ComfyUI backend.
• Local Ollama and LM Studio chat calls do not require provider keys.
• Ollama Web Search requires its own Ollama account API key only when the web search checkbox is enabled.

Useful Links

• ComfyUI
• OpenAI models
• Anthropic models
• Google Gemini models
• Groq models
• Mistral models
• Ollama Web Search
• Ollama OpenAI compatibility
• LM Studio OpenAI compatibility

Changelog

v1.2.1

• Added multi-chat sessions with persistent history.
• Added streaming responses through /llm-assistant/chat/stream.
• Added generation cancel support.
• Added image upload and provider payload adapters for vision-capable models.
• Added Ollama Web Search context injection.
• Added Prompt Enhancer tab and insertion into selected ComfyUI text/prompt widgets.
• Added Markdown rendering.
• Added thinking indicator with animated dots.
• Added voice input via Web Speech API.
• Updated provider defaults and model suggestions.
• Restyled the UI to better match ComfyUI.

v1.1.0

• Added Mistral API.
• Added Ollama.
• Added LM Studio.
• Added language selection: Russian, English, Ukrainian.
• Added beginner course generator.

v1.0.0

• First release with sidebar chat, quick actions, model/workflow browsing, and workflow helpers.

License

MIT License. See LICENSE.