Use AdaL with Local Models (Ollama)
Run AI models entirely on your own machine — no API key, no cloud costs, no data leaving your device. AdaL integrates with Ollama to let you use open-source models locally.
Local model support is in preview. Some features (e.g., image input) are not yet available for local models. Local models are smaller than cloud models and may struggle with complex tasks — if that happens, switch to a cloud model with /model.
Prerequisites
-
Install Ollama — download from ollama or:
curl -fsSL https://ollama.ai/install.sh | sh -
Start the Ollama server:
ollama serve -
Pull a model:
ollama pull gpt-oss:20b # good starting point
Selecting a Local Model in AdaL
Once Ollama is running with at least one model pulled, open the model selector:
/model
Scroll to the Ollama section at the bottom. Select your model — AdaL automatically routes to your local Ollama instance.
Supported Models
AdaL ships with pre-configured local model profiles:
| Display Name | Ollama Tag | Context | Best For |
|---|---|---|---|
| GPT-OSS 20B | gpt-oss:20b | 256K | General coding, reasoning |
| Qwen3-Coder 30B | qwen3-coder:30b | 128K | Code generation, refactoring |
You can pull any Ollama-compatible model and it will be available for selection.
Hardware Requirements
| Model Size | Minimum VRAM | Recommended |
|---|---|---|
| 7B | 8 GB | 16 GB |
| 14B | 16 GB | 24 GB |
| 30B | 32 GB | 48 GB |
Larger models produce better results but require more memory. On Apple Silicon Macs, unified memory is used — an M3 Max (36–128 GB) handles 30B models well.
Limitations (Preview)
- No image input — local models are text-only
- Slower response — depends on your hardware
- Tool calling quality varies by model
Troubleshooting
No reponses?
Make sure ollama serve is running and you have pulled the model via ollama pull <model> before launching AdaL. Check with:
curl http://localhost:11434/api/tags
Related: Models & Pricing