Jev directory
CommunityJev Tools & CLIs
Ollaya Decision Runtime
Ollaya's site for its local decision-model runtime: one binary pulls open checkpoints and serves them on TypeSafe-compatible endpoints, now at Ollaya 0.8.0 with an app and command line for macOS, Windows, Linux, and Docker.
- Reading time
- 2 min
- Published by
- Community
- Author
- Mert Cobanov
- Last verified
- Oct 2026
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- Ollaya 0.8.0 serves /v1/systemone, /v1/decisions, and /v1/models; the official TypeSafe Python SDK 0.7.1 works by changing TYPESAFE_BASE_URL.
- The site lists 21 model families: encoders like laya, nli, and gliclass plus decoders like winnow, kev, decider, and nimble.
- On Bespoke Labs' public benchmark over 3,880 human-labeled questions, winnow:12b posts 0.773 accuracy with ECE 0.141 at 60 ms a question.
- Same Nimble weights: ECE 0.022 on Ollaya against 0.122 on Ollama 0.35, which the site summarizes as 5.5x lower calibration error.
- winnow:e4b scores 0.722 at 89 ms for five questions on an RTX 4090 against hosted Jev's 0.738; the site lists 1.3k GitHub stars.
Watch out
Apache-2.0 runtime; each model keeps its own licence, and weights stay in the authors' Hugging Face repositories pinned by commit and sha256. Accuracy and latency figures are the project's own measurements on named hardware, with hosted-Jev numbers pulled from third-party benchmarks that include network time, so compare orders of magnitude only.
Questions about this entry
Short answers drawn from the linked project or page.
An independent open-source runtime that runs decision models locally: one binary pulls open weights and serves TypeSafe-shaped /v1/systemone, /v1/decisions, and /v1/models endpoints, with a desktop app and command line for macOS, Windows, Linux, and Docker.
Set TYPESAFE_BASE_URL to http://localhost:11435, TYPESAFE_API_KEY to any value, and TYPESAFE_DEFAULT_MODEL to a served model such as winnow:e4b; the official TypeSafe Python SDK 0.7.1 works unchanged.
The site's table puts winnow:e4b at 0.722 accuracy and 89 ms for a five-question request on an RTX 4090, against 0.738 and 236-276 ms for TypeSafe's hosted Jev; the hosted-Jev numbers come from third-party benchmarks with network included.
Side by side on the same RTX 5090 over 3,880 labeled questions, Ollaya leads on accuracy and calibration: winnow:12b scores 0.773 with ECE 0.141, while Ollama's best is 0.749 on Nimble with ECE 0.122, against 0.022 for the same weights on Ollaya.
There are no per-token fees and no metering; it runs as many decisions as the hardware handles. The runtime is Apache-2.0, and every model keeps its author's licence.
21 families, split between encoders like laya, nli, gliclass, von, and decima and decoders like winnow, clef, kev, decider, nimble, and snap. Small encoders answer in milliseconds on a CPU.
No. The server listens on 127.0.0.1 by default, runs on ONNX Runtime on CPU or an NVIDIA GPU, and reads weights from the authors' Hugging Face repositories without re-hosting them.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Standard One SH (StandardThinking)
LinkedIn: What multimodal decision models actually change, and what they do not.
LinkedInJevDirectory.org
Hacker News: Decision Index 0.2.1 scores open reproductions of Jev on a frozen suite.
jackbravo4 pointsDecision Index 0.2.1 – 70 open reproductions of TypeSafe Jev's Decision ModelRead the thread on Hacker NewsHacker News: Pointer to a DiffusionGemma plus vLLM tutorial for a Jev-shaped classifier.
1 pointHow to build a Jev-style classifier with DiffusionGemma and vLLMRead the thread on Hacker NewsMore Jev resources like this
A local daemon that pulls open decision models and serves them on TypeSafe's /v1/systemone, so a Jev client switches hosts with TYPESAFE_BASE_URL.
Jev Models & Reimplementations#community#rust#open-models
Community
ollaya-dev/ollayaRust
ollaya-dev/ollayaRustAn Ollama-style local runner for System One decision models from Hugging Face: pull, serve and query them behind the same /v1/systemone API as TypeSafe's Jev, with a terminal model manager and a browser playground.
Jev Tools & CLIs#community#catalog-import#madewithjev
Community
nvkudva/ollajevPython
nvkudva/ollajevPythonThe Ollama build of iapp's OpenThai-SystemOne: a 0.8B Thai and English decision model whose v0.3 weights are fine-tuned for Ollama's /v1/systemone letter readout, reporting a 74.3 macro across 13 public subsets for Q8_0.
Jev Models & Reimplementations#community#open-models#local
Community
A local server that reads next-token probabilities from a fine-tuned Spark model and returns Choice, Score, and Noul answers with zero generated tokens.
Jev Models & Reimplementations#community#python#open-models
Community
Rizzo-AI-Academy/rizzo-flowPython
Rizzo-AI-Academy/rizzo-flowPythonA drop-in proxy that distills a repeated Jev classification call into a local model as it runs: it learns from Jev's own probability distributions, then answers most requests itself while uncertain ones and a 2% audit slice keep going to Jev.
Jev Tools & CLIs#community#python#distillation
Community
tomerglick57/JevstillerPython
tomerglick57/JevstillerPythonA proof-of-concept HTTP service for constrained decisions with a local MLX model: send text or JSON plus typed choice, score, or noul questions and it returns probabilities over the answers you declared — no generated text, no parsed model JSON.
Jev Tools & CLIs#community#github#python
Community
snellingio/system-onePython
snellingio/system-onePythonFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.