Local Models
Coming soonChoose where your model runs.
Run supported models locally for code review, internal drafts, and experimentation. Keep the runtime and data path in view. Beta.
Local Models guide
Uses
Bring local models to real work.
For builders, researchers, and writers who want more direct control over model execution, where runtime support allows it.
Review code
Explore a local review of unreleased code, then verify the findings.
Draft internal work
Refine notes and plans before deciding what to share.
Test open weights
Compare model behavior on summaries, extraction, and other repeated tasks.
Boundaries
Understand what stays local.
Local execution does not make the whole Ethen workspace offline. Account services, metadata, history, configuration, and connected tools may still use cloud services. Check the runtime and data path before adding sensitive context.
Model Choice
Match the model to the task.
Local execution depends on runtime support, model availability, and machine capability. Open weights describe model access under a license; they do not determine where execution happens.
- Model
- qwen2.5-coder:7b
- Family
- qwen2
- Parameter size
- 7.6B
- Quantization
- Q4_K_M
- License
- Shown from the model's Modelfile
- Capabilities
- Reported by the local runtime
Field names from the desktop panel; values are an example.
Workflow
Review before you share.
Check the setup
Confirm the supported runtime, model, and machine requirements.
Choose the context
Bring only material appropriate for the configured data path.
Run and review
Inspect the output, assumptions, and any verification needed.
Decide what leaves
Review sensitive results before sharing or moving them into connected services.
Local vs hosted
What runs where, exactly.
Local means loopback
Loopback-only Ollama: the Code daemon talks to Ollama on your machine only — status, list, show, approval-gated pull with progress, and resume-safe cancel.
No hosted execution
There is no hosted tier and no hosted fallback. Pulls cross the network to fetch weights; inference stays on-device.
The workspace stays connected
Local execution does not make the Ethen workspace offline. Account services, metadata, history, and connected tools may still use cloud services.
Ollama only, in the Code path
llama.cpp and LM Studio adapters exist only in the frozen monolith — not in the Code branch this page describes.
Inside the desktop panels.
The models panel reports the runtime and what is installed; pulling a model asks for typed consent and shows each stage. Both are recreated here from the console components until the desktop ships.
| Model | Family | Parameter size | Quantization |
|---|---|---|---|
| qwen2.5-coder:7b | qwen2 | 7.6B | Q4_K_M |
| llama3.2:3b | llama | 3.2B | Q4_K_M |
| nomic-embed-text | nomic-bert | 137M | F16 |
Pull qwen2.5-coder:7b
Weights download over the network; inference stays on this machine.
- Pulling manifest…
- Downloading layers…
- Verifying digest…
- Writing manifest…
FAQ
Does Local Models work entirely offline?
Local model execution can still sit inside a cloud-connected workspace. Network needs depend on the workflow and configuration.
What are open-weight models?
Models whose weights are available for direct use under their applicable licenses. Local execution also requires a compatible runtime and suitable hardware.
Can I use sensitive code or notes?
Where the supported setup fits your requirements. Check the runtime, data path, and connected workspace services before adding sensitive material.
Will local models cost less or perform better?
That depends on the model, hardware, and task. Test the setup against your needs.
Start with your local setup.
Local Models is in beta and ships with the Code desktop. Check the requirements, then choose a supported model.