Ollama & Offline Mode
Run HelixMind fully offline with Ollama. no API key, no data leaves your machine.
Ollama & Offline Mode
HelixMind works fully offline with Ollama. No API keys, no data leaves your machine.
Setup
- Install Ollama from ollama.ai
- Start the server:
ollama serve - Configure HelixMind:
helixmind config set provider ollama
helixmind config set model qwen2.5-coder:32b # optional
HelixMind auto-detects Ollama at http://localhost:11434.
Recommended Models
| Model | Size | Quality | Speed |
|---|---|---|---|
qwen2.5-coder:32b | 19GB | Excellent | Moderate |
qwen3-coder:30b | 17GB | Excellent | Moderate |
qwen2.5-coder:14b | 8.5GB | Good | Fast |
qwen2.5-coder:7b | 4.4GB | Decent | Very Fast |
deepseek-r1:14b | 8.5GB | Good (reasoning) | Moderate |
For best results, use 14B+ parameters. 32B recommended for complex tasks.
Model Management
| Method | Command |
|---|---|
| In chat | /local: list, pull, manage models |
| Switch model | /model: interactive picker |
| CLI pull | ollama pull qwen2.5-coder:32b |
| CLI list | ollama list |
What Works Offline
| Feature | Status |
|---|---|
| Interactive Chat | ✅ Full |
| Agent Tools (14) | ✅ Full |
| Spiral Memory | ✅ Full |
| Brain Visualization | ✅ Full |
| Validation Matrix | ✅ Full (uses same model) |
| Checkpoints & Undo | ✅ Full |
| Sessions | ✅ Full |
| MCP Server | ✅ Full |
| Web Knowledge Enricher | ❌ Disabled (requires internet) |
Hardware Requirements
| Model Size | RAM | GPU VRAM |
|---|---|---|
| 7B | 8GB | 4GB |
| 14B | 16GB | 8GB |
| 32B | 32GB | 16GB |
Ollama auto-uses GPU if available (NVIDIA CUDA, Apple Metal). Store models on SSD for faster loading.
Privacy
- Zero data leaves your machine
- No API key required
- No telemetry or analytics
- Ideal for corporate environments, sensitive code, air-gapped networks