Local-first cost-aware LLM routing

Use the right model only when the work actually needs it.

muLLM routes simple queries to local models, cache, and groundtruth first, then escalates to cloud models only when complexity or policy requires it.

Download View source

Private by default

Local routing, local cache, local setup, and explicit cloud provider toggles.

Budget guarded

Per-query estimates, hard caps, cancellation, and audit events are first-class surfaces.

Modular

Core router stays small. Studio, benchmarks, and coding orchestration are opt-in modules.

localfirst route
cloudonly when needed
auditevents and exports
Apache-2.0license

Install the 1.0 release candidate from source

git clone https://github.com/mybrainrunslinux/mullm.git
cd mullm
python -m pip install .
mullm-server

PyPI publication is pending. Core installs do not require constraints.txt. Open /factory to follow projects and answer Need Human decisions.