<free/>

field notes / local AI

Open-source models for local coding

Updated November 4, 2026

01

Start here

Explore Nimbus model releases and downloads

All guides

Choose local coding models for notch: check licenses, hardware, tool support and real project tests. Understand open source versus open weights.

02

What can I run with notch?

notch is the FreeSyntax coding workspace and terminal assistant for local AI, with optional cloud models. The Windows preview combines an editor, project files, terminal and assistant. The CLI supports terminal-based work. Local model execution depends on your installed runner and a compatible model; downloading weights alone does not configure a coding assistant.

03

Open source and open weights describe different things

A downloadable model is not automatically open source. Read the release’s license, model card, training-code and data documentation. The Open Source Initiative definition considers permissions and the information needed to study and modify the system, beyond availability of weights. Third-party model terms are separate from the coding tool’s terms. We do not claim every model compatible with notch meets that definition.

04

Choose a model for your actual machine

Check the runner’s supported model format, available memory, context settings and tool-calling support. Larger context and concurrent applications can increase memory use. Quantized files can reduce memory needs, with quality and compatibility tradeoffs. Avoid treating a parameter count or download size as a guarantee that a model fits.

05

Use a small project task as your acceptance test

Start with a repository you can safely change. Ask the assistant to explain one function, then make one narrow fix. Review the diff and run the project’s existing tests. Record the model identifier, quantization, runner and settings with the result. The landing demos show selection, download and simple tasks; they are demonstrations, not comparative benchmark claims.

06

Check rights and behavior before adoption

Confirm the release permits your intended use, redistribution and modifications. Evaluate code quality, license obligations and whether its tool calls work in your runner. Check the original model card rather than relying on an “open” label in a directory. Keep a working configuration so upgrades can be compared against the same task.

07

Original sources