README
Bforartists Coworker
AI-powered 3D creation, built right into Blender.
Coworker is a self-contained Bforartists / Blender add-on that puts a local AI assistant directly in your 3D viewport. No external tools, no server setup, no Python environment wrangling -- install the add-on, pick a model, start chatting.
The coworker can inspect your scene, write and execute Blender Python code, place objects, set up materials, configure lighting, animate, render, and more -- all through natural conversation.
🚧 WORK IN PROGRESS — This is an active fork under heavy development. See '_misc' for planned features and known gaps.
⚠️ Windows tested only for now
Why Coworker?
- Zero setup -- Download llama-server with one click, pick a model preset, start talking. No terminal, no Docker, no API keys required for local use.
- Runs locally -- Your scenes never leave your machine. Models run on your own GPU via llama.cpp. No cloud dependency.
- Actually useful -- 30+ tools let the coworker inspect scenes, execute code, render viewport screenshots, browse assets, and modify your project in real time.
- Works with what you have -- Already downloaded a GGUF model? Coworker finds it automatically. Has a HuggingFace token? Paste any model URL to download.
- Remote API too -- Prefer a cloud model? OpenRouter, OpenAI, and Anthropic are one dropdown away with automatic URL configuration.
Quick Start
- Install -- Edit > Preferences > Add-ons > Install from Disk > select the .zip
- Download llama-server -- Click the one-click download button in preferences (auto-detects CUDA/Vulkan/CPU)
- Pick a model -- Choose from curated presets organized by your hardware:
- Flagship (24 GB+ VRAM): Qwen3.8-27B, Fable Fusion 27B, Nail 35B
- Mid-Range (16 GB): GPT-OSS 20B (default), Qwen3.8-27B Q4
- Lightweight (<=8 GB): Gemma 4 E4B, Qwen3.5-9B variants
- Download & Start -- One click downloads the model and launches the server
- Chat -- Open the N-panel in the 3D Viewport, find the Coworker tab, and start creating
What Can It Do?
| Area | Examples |
|---|---|
| Modeling | "Create a stonehenge with 8 pillars in a circle" |
| Materials | "Apply a weathered stone material to all pillars" |
| Lighting | "Set up dramatic three-point lighting" |
| Animation | "Make the ball bounce with squash and stretch" |
| Rendering | "Render a thumbnail from the active camera" |
| Scene Management | "Organize everything into named collections" |
| Asset Downloads | "Download a forest HDRI from Polyhaven" |
| Vision | "Look at the viewport and reframe the camera for a hero shot" (vision models) |
The AI writes and executes Blender Python code in real time -- you see changes happen live.
Download
| Latest Release | Download v1.1.37 |
| All Releases | GitHub Releases |
| Source Code | Clone this repo |
Requirements: Blender 5.1+ (or Bforartists equivalent), ~4-28 GB RAM depending on model.
Get Involved
Coworker is open source under GPL-3.0. Contributions are welcome -- whether it is bug reports, feature ideas, model testing, documentation, or code.
- Report issues -- GitHub Issues
- Discuss ideas -- GitHub Discussions
- Read the docs -- Wiki
- See what changed -- CHANGELOG
How It Works
You type in the Chat UI in the interface
|
Agent Controller talks to Local LLM (AI) or Remote API or External Harness
|
MCP Server receives tool calls
|
Blender executes the code
|
You see the result in your blend file, viewport or editors
Everything runs inside Bforartists / Blender -- no external processes to manage, no ports to configure.
License
GPL-3.0-or-later -- see LICENSE.
Contributors
221 commits
Similar repos
Languages
Python
99.7%
README
Bforartists Coworker
AI-powered 3D creation, built right into Blender.
Coworker is a self-contained Bforartists / Blender add-on that puts a local AI assistant directly in your 3D viewport. No external tools, no server setup, no Python environment wrangling -- install the add-on, pick a model, start chatting.
The coworker can inspect your scene, write and execute Blender Python code, place objects, set up materials, configure lighting, animate, render, and more -- all through natural conversation.
🚧 WORK IN PROGRESS — This is an active fork under heavy development. See '_misc' for planned features and known gaps.
⚠️ Windows tested only for now
Why Coworker?
- Zero setup -- Download llama-server with one click, pick a model preset, start talking. No terminal, no Docker, no API keys required for local use.
- Runs locally -- Your scenes never leave your machine. Models run on your own GPU via llama.cpp. No cloud dependency.
- Actually useful -- 30+ tools let the coworker inspect scenes, execute code, render viewport screenshots, browse assets, and modify your project in real time.
- Works with what you have -- Already downloaded a GGUF model? Coworker finds it automatically. Has a HuggingFace token? Paste any model URL to download.
- Remote API too -- Prefer a cloud model? OpenRouter, OpenAI, and Anthropic are one dropdown away with automatic URL configuration.
Quick Start
- Install -- Edit > Preferences > Add-ons > Install from Disk > select the .zip
- Download llama-server -- Click the one-click download button in preferences (auto-detects CUDA/Vulkan/CPU)
- Pick a model -- Choose from curated presets organized by your hardware:
- Flagship (24 GB+ VRAM): Qwen3.8-27B, Fable Fusion 27B, Nail 35B
- Mid-Range (16 GB): GPT-OSS 20B (default), Qwen3.8-27B Q4
- Lightweight (<=8 GB): Gemma 4 E4B, Qwen3.5-9B variants
- Download & Start -- One click downloads the model and launches the server
- Chat -- Open the N-panel in the 3D Viewport, find the Coworker tab, and start creating
What Can It Do?
| Area | Examples |
|---|---|
| Modeling | "Create a stonehenge with 8 pillars in a circle" |
| Materials | "Apply a weathered stone material to all pillars" |
| Lighting | "Set up dramatic three-point lighting" |
| Animation | "Make the ball bounce with squash and stretch" |
| Rendering | "Render a thumbnail from the active camera" |
| Scene Management | "Organize everything into named collections" |
| Asset Downloads | "Download a forest HDRI from Polyhaven" |
| Vision | "Look at the viewport and reframe the camera for a hero shot" (vision models) |
The AI writes and executes Blender Python code in real time -- you see changes happen live.
Download
| Latest Release | Download v1.1.37 |
| All Releases | GitHub Releases |
| Source Code | Clone this repo |
Requirements: Blender 5.1+ (or Bforartists equivalent), ~4-28 GB RAM depending on model.
Get Involved
Coworker is open source under GPL-3.0. Contributions are welcome -- whether it is bug reports, feature ideas, model testing, documentation, or code.
- Report issues -- GitHub Issues
- Discuss ideas -- GitHub Discussions
- Read the docs -- Wiki
- See what changed -- CHANGELOG
How It Works
You type in the Chat UI in the interface
|
Agent Controller talks to Local LLM (AI) or Remote API or External Harness
|
MCP Server receives tool calls
|
Blender executes the code
|
You see the result in your blend file, viewport or editors
Everything runs inside Bforartists / Blender -- no external processes to manage, no ports to configure.
License
GPL-3.0-or-later -- see LICENSE.
Similar repos
Contributors
221 commits
Languages
Python
99.7%