Built on MLX
An inference engine built around Apple's MLX framework and its Apple-silicon ecosystem.
Foundation · Product 01
IN DEVELOPMENTForge.Memory-efficient inference. A more connected model workflow.
Forge is Foundation's upcoming MLX-based inference engine, designed to make model execution more memory-efficient. The broader workflow connects preparation, training, inference, benchmark evaluation, and export.
01
Prepare
02
Train
03
Infer
04
Evaluate
05
Export
Designed around efficiency
Less friction between a model and the work it needs to do.
Forge is in development. Its focus is a memory-efficient inference engine and a more coherent path through the stages of model development.
Memory efficiency first
Designed to reduce the memory burden of model execution on compatible local machines.
One connected process
Bring preparation, training, inference, benchmark evaluation, and export into a single development path.
The Forge workflow
From preparation to export.
Prepare
Bring the model and its required assets into a clear, repeatable workflow.
Train
Keep model training connected to the wider development process.
Infer
Run models through an MLX-based inference engine designed with memory efficiency in mind.
Evaluate
Use benchmark evaluation to check model behavior and compare results.
Export
Prepare the model for the next stage after development and evaluation.
Forge · In development
Make every stage work together.
Forge is under development. Contact Foundation to discuss the direction of the project.
Get in touch