Jev vs Ollaya
Ollaya is not affiliated with Ollama, and not affiliated with TypeSafe.
Jev is TypeSafe’s hosted System One API. Ollaya is an independent local runtime for open decision models with a TypeSafe-compatible HTTP API.
This unofficial note describes documented interfaces only. It does not endorse either product.
What each project is
Jev is TypeSafe’s hosted System One model: state + typed questions in, structured answers out. Official calls use POST https://api.typesafe.ai/v1/systemone and the TypeSafe SDKs. The Python SDK reads TYPESAFE_API_KEY and defaults to jev-latest.
Ollaya is a local server and CLI that pulls and serves open decision models. Its FAQ says it borrows Ollama’s experience (one binary, pull / run / serve, Modelfiles, a local REST API) and applies that pattern to decision models instead of generative language models. That is a UX analogy, not an affiliation.
Ollaya’s documented job is to run those open models on your hardware and expose TypeSafe-shaped endpoints. Compatibility is documented as covering the API, not TypeSafe’s model.
Documented surfaces
| TypeSafe Jev | Ollaya | |
|---|---|---|
| Product kind | Hosted System One model | Local runtime + model library |
| Official site | typesafe.ai · docs | ollaya.dev |
| Decide endpoint | POST /v1/systemone on api.typesafe.ai | POST /v1/systemone (and /v1/decisions) on 127.0.0.1:11435 by default; native POST /api/decide |
| List models | TypeSafe account models | GET /v1/models — models pulled onto this machine |
| Default listen | Hosted | 127.0.0.1:11435 (next to Ollama’s default 11434, so both can run) |
Ollaya documents that the TypeSafe Python SDK 0.7.1 works unchanged when you set:
export TYPESAFE_BASE_URL=http://localhost:11435
export TYPESAFE_API_KEY=local
export TYPESAFE_DEFAULT_MODEL=layalocal is a placeholder. Ollaya accepts any key unless the server sets OLLAYA_API_KEY. If you omit TYPESAFE_DEFAULT_MODEL, the SDK still sends its default jev-latest, which Ollaya will not have pulled.
Models Ollaya lists
Ollaya’s FAQ lists open families you can run (each with its own license and authors), including laya (Convai Innovations), decider (Mapika), nli (Moritz Laurer), gliclass (Knowledgator), kev (Jared Palmer), decision (vLLM Semantic Router contributors), qwen3guard (Qwen team), von (Victor Hugo Panisa), winnow (EldanRing), and jevk5 (alibiserikbay). Weights are pulled from the authors’ Hugging Face repos, pinned and checked against sha256. Ollaya states it does not re-host those weights.
Platforms Ollaya documents
- Linux x86-64 and ARM64 with glibc 2.38 or newer
- macOS on Apple silicon
- Windows 10 and 11 on 64-bit x86: desktop app on CPU; command line uses an NVIDIA GPU when present (FAQ). WSL 2 with the Linux installer is optional.
- Docker image
ghcr.io/ollaya-dev/ollaya(:cudafor NVIDIA)
A GPU is not required. On x86-64 Linux and Windows, the command line uses an NVIDIA GPU with driver R580 or newer (CUDA 13) when one is present (FAQ). WSL 2 is optional.
Install paths Ollaya publishes: desktop app; curl -fsSL https://ollaya.dev/install.sh | sh on Linux/macOS; irm https://ollaya.dev/install.ps1 | iex on Windows; or the Docker image.