Open source · 2026
Local Lattice

Try it
An illustration of how a request is routed. Which model fills each role is up to you.
Ask for
Mode
Your app
model: "role:coder"LatticeOne model
- MLXon-device
- LM Studiolocal server
- Cloud keyyour API key
How it came together
- Problem
- Every local model server and cloud provider names its models differently, so apps hard-code one and break when you switch.
- Constraint
- Apps already speak the OpenAI API, so the fix couldn’t ask them to change.
- Decision
- A gateway that speaks the OpenAI API but takes a role like coder or reasoner instead of a model name, then routes to MLX, LM Studio, or a cloud key. It can also send one question to several models and have them vote, chain, or debate.
- Outcome
- Published on PyPI as local-lattice, so it’s one pip install.
At a glance
- Role
- Solo · designed and built
- Shipped
- 2026 · open source
- Built with
pip install local-lattice