OmniJev
OmniJev is an open research organization/project focused on Jev-like decision models and the infrastructure around them. Its Flide project provides fast, exact inference for decision models, while the broader OmniJev work includes multimodal models, datasets, benchmarks, and related open-source projects.
At a Glance
- AI-agent and decision-intelligence developers
- Researchers working on LLM inference, Jev/System One models and calibrated categorical reasoning
- Teams needing high-throughput, low-latency classification or triage
- Computer-use and GUI-agent developers
- +1 more
AI Tools by OmniJev
(1)
Flide
Open Source Decision Model Inference
Discussions
No discussions yet
Be the first to start a discussion about OmniJev
Latest News
Flide 0.1.11 added an email-triage demo processing 10,000 emails with eight questions each in 28 seconds on one H200.
Flide 0.1.10 added Ainvest-Decision, Darwin-27B-RSI-Decision and KnowLine-31B-Gen1.5; KnowLine gained image support.
Flide 0.1.9 added pplx-decider-v1.1-27b and JEV-35B, and the /v1/systemone API began accepting several requests in one body.
Flide 0.1.8 added AutoTrust and Clef model families plus drex-max-v1.
Products & Services
Open-source inference infrastructure and Python/PyPI package for Jev-style decision models. It uses exact state forking and serves the Jev /v1/systemone API; current documentation supports Qwen3.5, Gemma 4, OneJev, Open-Jev and many other decision-model families.
A multimodal System One decision model based on Qwen, available in approximately 0.8B, 4B, 9B and 27B sizes. It accepts text, images and video and returns calibrated probabilities for typed questions in one forward pass.
Open dataset for OneJev; the OneJev repository describes 94,707 released examples.
A 0.8B JEV-like multimodal model for playing GUI games directly from raw pixels.
Market Position
Flide positions itself as an inference engine specialized for Jev-style decision workloads rather than general text generation. Its official benchmark claims up to 9.5x lower latency than vLLM and SGLang on the tested workloads, with exact state forking and closer agreement to a float32 reference; the OneJev repository also contrasts its serving path with qev and llama.cpp. The relevant comparison set is therefore vLLM, SGLang, qev and llama.cpp, with Flide differentiated by shared-state decision execution and exactness.
Business Model
Revenue Model
Open-source software and model distribution: Flide is published on PyPI under Apache-2.0 and the associated models, datasets and repositories are publicly available. The official materials present no paid SaaS or enterprise pricing model.
Pricing Tiers
Installable from PyPI; Apache-2.0 licensed.
Target Markets
- AI-agent and decision-intelligence developers
- Researchers working on LLM inference, Jev/System One models and calibrated categorical reasoning
- Teams needing high-throughput, low-latency classification or triage
- Computer-use and GUI-agent developers
- Open-source AI and multimodal-model users with NVIDIA GPUs
- Low-latency categorical decision inference for LLM agent systems
- Agent workflows that ask many questions about one shared state
- Email triage and multi-label classification
- GUI and browser-agent perception and action selection
- Multimodal decision making over text, screenshots, photos and video
- Research and benchmarking of Jev/System One decision models