Run AFM-D decision models with Ollaya - Aria Compute

Ollaya runs open decision models locally the way Ollama runs LLMs. A decision model takes a text or JSON state plus typed questions (choice, score, noul) and returns calibrated probabilities in one forward pass. It never generates chat text. Aria Compute ships AFM-D checkpoints that Ollaya can pull by catalog name:

Catalog	Role
`afm-de`	AFM-D Encoder (ModernBERT + MASK option head)
`afm-dd`	AFM-D Decoder (MiniCPM5-2B SemIf option-letter logits)

Weights come from Hugging Face (ariacompute/*) or ModelScope (AriaCompute/*). Set OLLAYA_HUB when you need the China mirror. See AFM-D for product positioning.

Install and pull

curl -fsSL https://ollaya.dev/install.sh | sh
ollaya pull afm-de
ollaya pull afm-dd

Serve System One

ollaya serve

Ollaya speaks TypeSafe-compatible POST /v1/systemone (default listen is documented in the Ollaya project; typically http://localhost:11435). Point clients that already use System One at that base URL.

When to use Ollaya vs Aria Engine

	Aria Engine	Ollaya
Focus	Official AFM System One runtime (`aria-engine`)	Ollama-like UX for many open decision models
AFM-D	Native `afm-de` / `afm-dd` tracks	Catalog pull + local daemon
Wire format	`POST /v1/systemone`	Same System One shape

Use Engine when you want the Aria Compute release binary and FFI. Use Ollaya when you want a local decision-model daemon with pull/list/run ergonomics.

Next steps

AFM-D concepts
Engine runtime
Playground for interactive trials on the Aria Compute site