Theme
AI Resources
OpenJev
OpenJev is an open-source server for asking bounded yes-or-no, choice, and score questions and receiving probabilities instead of generated prose.
It runs DiffusionGemma through a Jev-compatible API, so builders can use structured decisions for routing, filtering, scoring, or gates without generating an answer and parsing it back into a fixed shape. Use this as a first read, not a recommendation. Open the original project before trusting details like terms, limits, privacy, cost, setup, or safety.
What it is
A server for small, typed decisions
An application sends a state plus named questions. OpenJev returns probabilities and confidence for yes-or-no, choice, and score outputs through an HTTP API.
Why it stands out
It reads answer probabilities without free-text output
OpenJev places the answer structure onto DiffusionGemma's denoising canvas and reads the probability distribution at each answer slot. That keeps responses inside the supplied schema.
Availability
Self-hosted or available through Codiv
The repository provides source code, tests, Docker setup, and a prebuilt image. A hosted endpoint is also documented by Codiv, while self-hosting the listed checkpoint requires an NVIDIA GPU with substantial memory.
Why it matters
What makes it useful
Many agent steps are narrow decisions rather than writing tasks: choose a route, score urgency, or decide whether a condition is met. OpenJev makes that pattern inspectable as a separate service with explicit options and probability outputs.
What to know
Where it fits
It fits builders testing typed decision steps inside agents, queues, classifiers, or policy workflows. The Jev-compatible interface also gives TypeSafe SDK users an independently maintained endpoint they can inspect or run themselves.
Notable points
What stands out
OpenJev is a new independent project, not TypeSafe AI's model or an official TypeSafe product. The repository does not currently list a formal GitHub release and depends on an unmerged vLLM change, so the project pins its own vLLM branch and documents fixes around that path.
Before using
What to review
Whether the workload really needs a bounded decision with supplied options rather than generated text or a general-purpose classifier.
Whether the self-hosted hardware fits: the documented NVFP4 setup calls for an NVIDIA GPU with at least 24 GB of memory and downloads roughly 18 GB of model weights.
Whether the project's early maturity and provisional vLLM dependency fit the workflow. The repository recommends evaluating answer quality on your own tasks before relying on it.
Whether to run it locally or use Codiv's hosted endpoint, where prompts and credentials cross that external service boundary.
Reader fit
Who may find it relevant
Builders experimenting with typed decisions for agent routing, triage, filtering, or scoring.
Teams that want an inspectable Jev-compatible server and can evaluate its results against their own tasks.
Less relevant for readers who want a mature production service, lightweight CPU setup, or a general chatbot that mainly writes free-form responses.
Editorial note
Why LifeHubber lists it
LifeHubber lists OpenJev because it turns a common but easy-to-hide part of an AI system—small routing and scoring decisions—into an explicit API with named choices and probabilities. That helps builders inspect whether a separate decision service fits better than asking a general chat model for structured text.
Source links
Source materials
Reader note
Before relying on this entry
LifeHubber lists entries to help readers inspect AI projects, not to endorse them or prove they are safe, suitable, accurate, maintained, or right for a specific use. We do not verify every entry in depth. Before relying on anything listed, review the original materials, terms, privacy practices, limits, and risks that matter for your situation.
More in Ecosystem
Keep browsing this category
Explore more AI ecosystem resources.
LEANN
StarTrail-org/LEANN
A local vector index for semantic search and personal RAG that reduces stored embeddings through selective recomputation, with Python, CLI, and MCP routes.
exo
exo-explore/exo
A local inference system that joins Apple Silicon Macs or Linux machines into a cluster, automatically places and shards models across available hardware, and exposes a dashboard plus several compatible APIs.
Apple Core AI Models
apple/coreai-models
An Apple repository for building on-device AI around Core AI, with export recipes for supported open-source models, Python primitives for PyTorch authoring, Swift runtime utilities for macOS and iOS apps, a model catalog, CLI paths, and coding-agent skills.
For project maintainers
Listed here? You can use the badge.
If you maintain a project with a current LifeHubber listing, you may add the optional “Listed on LifeHubber AI Resources” badge to its README, docs, or website. No introduction or permission request is needed.
