Back to Software
AI agent building blocksAvailable

MSM — Multi Small Models

An open standard that replaces one giant model with a pipeline of small, specialised ones.

Giant models are expensive per call, slow, hard to audit, impractical to self-host — and English-first by birth. For Arabic-speaking Gulf markets that is not a detail: a model that treats Arabic as a secondary language cannot serve an Arabic-speaking customer well, whatever its parameter count.

The usual answer is to pile engineering on top of the giant model. Ours is different: a pipeline of layers defined by strict contracts — translation, classification, orchestration, generation, validation, then outbound translation. Any model satisfying a layer's contract is valid there, and swapping it is one line in a manifest, not a code change.

The practical result is that every call returns a full per-layer trace: what each layer said, with what confidence, in how long. A system you can audit, not a black box.

Highlights

  • Layers are typed contracts — any model meeting the contract fits
  • Swapping a model is one line in a YAML manifest
  • A full trace per layer: output, confidence, latency
  • Runs on open models locally — no single-provider lock-in

Want one of these in your business?

Whether it's an off-the-shelf product or a custom build on our engines, we'll help you deploy it.

Get in Touch