localcode 0.2.11rc8

High-performance AI coding on consumer hardware.

🏠 LocalCode

High-performance AI coding on consumer hardware.
No cloud, no API keys, no data leaving your machine.

Install

pip install localcode

Run

cd your-project
localcode

That's it. First launch builds the inference server and downloads the model (~5 min, one time). After that, startup is ~15 seconds.

What it does

• Reads and edits files - understands your codebase, makes surgical edits
• Runs commands - tests, builds, git, shell
• Searches code - by pattern, content, or semantic meaning
• Fast mode - for routine coding tasks
• Reasoning mode - deep thinking for complex multi-step problems
• Uses tools automatically - the model picks its own tools

> refactor the auth module to use JWT and make sure the tests pass

LocalCode reads the files, plans the refactor, edits the code, runs the tests, and fixes failures - all locally.

Why local?

We are building for a world of truly democratized AI - where everyone has access to powerful, personalized, prompt AI anywhere, on any device, and in any location. True empowered local-first AI. LocalCode is the first step toward that vision.

How LocalCode compares

	LocalCode	Claude Code	OpenCode	Codex CLI
Runtime	100% on-device	Cloud (Anthropic API)	Cloud (any provider)	Cloud (OpenAI API)
Privacy	Code never leaves your machine	Code sent to Anthropic	Code sent to provider	Code sent to OpenAI
Cost	Free forever	$100+/mo (Max) or API credits	API credits (varies)	Free (included with ChatGPT)
Internet required	No	Yes	Yes	Yes

Requirements

• Mac with Apple Silicon (M1/M2/M3/M4)
• 16GB RAM minimum
• Python 3.11+
• ~12GB free disk (10GB model + server)

How LocalCode works

LocalCode runs a custom llama.cpp fork with TurboQuant KV cache compression - a technique from Google's ICLR 2026 paper that we patched into llama.cpp for Apple Silicon. This compresses the KV cache 3.8x, fitting 32K context in 355 MiB on a 16GB MacBook.

The model (Gemma 4 26B-A4B) is a Mixture-of-Experts architecture - 25.2B total parameters but only 3.8B active per token. That's what makes 27 tok/s possible on a laptop.

Sponsors

If you'd like to sponsor LocalCode, reach out.

Contributing

See CONTRIBUTING.md.

License

Apache-2.0. See LICENSE.