Self-host a Laya API server

Official: laya-serve

From the core README:

pip install "laya[serve]"
LAYA_DEVICE=cuda LAYA_PRELOAD=1 laya-serve   # binds 0.0.0.0:8000, preloads all 3 checkpoints

curl -s localhost:8000/v1/systemone -H 'content-type: application/json' -d '{
  "state": {"body": "billed twice, refund please or we cancel"},
  "questions": {"dept": {"type": "choice", "instructions": "which team?",
                "criteria": {"billing": "refunds", "tech": "bugs"}}}
}'

Set LAYA_API_KEY to require a bearer token. Other settings are LAYA_HOST, LAYA_PORT, LAYA_MODELS and LAYA_THREADS. A Nix flake and a NixOS module are included.

ollaya

ollaya works like Ollama for decision models:

curl -fsSL https://ollaya.dev/install.sh | sh
ollaya run laya --preset triage "I was charged twice for my subscription this month and want a refund."

It serves /v1/systemone, and the README says the official TypeSafe SDK works unchanged with TYPESAFE_BASE_URL=http://localhost:11435.

Docker

LAYA SERVER adds a web console and API-key management:

docker run -d --name laya-server --init --restart unless-stopped \
  -p 8080:8080 \
  -v laya-data:/data \
  -e LAYA_ADMIN_USERNAME=admin \
  -e LAYA_ADMIN_PASSWORD='change-this-admin-password' \
  1panel/laya-server:latest

docker-laya publishes ghcr.io/chneau/laya for amd64 and arm64. Its README reports roughly 0.35 s per predict on CPU.

Native servers

laya.cpp (C++, CUDA, Vulkan and Core ML), sys1 (Rust on candle) and laya-apple (MLX plus Neural Engine) all serve /v1/systemone.

Differences from Jev

The core README lists three for client ports. Options share a token budget instead of Jev's 255-option cap. Every score level needs a description. confidence is computed differently, so a threshold tuned on Jev does not transfer; gate on answer_confidence instead. laya-rs uses /v1/system-one with a hyphen and is not a drop-in replacement.

Hosted Laya APIs

If you would rather not run a server at all, some services host Laya behind the same /v1/systemone wire format, among them layahost (EU, run by the makers of laya.tools) and Laya Studio (Switzerland). A one-click Railway template sits in between: still self-hosting, on Railway's machines. What to compare between them and your own server: Hosted Laya API: when not to self-host.