Organization repository on GitHub
A local, API-compatible replacement for Jev, TypeSafe's System One model
- Forks
- 4
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Models & Reimplementations
Lichen
A local /v1/systemone server that reads label probabilities from a GGUF chat model, with prompt repetition and a confidence shrink toward uniform.
- Stars
- 58
- Language
- Python
- License
- MIT
- Reading time
- 3 min
- Published by
- Community
- Author
- Mushroom-Systems
- Last verified
- Sep 2026
About this project
Lichen is a drop-in, API-compatible replacement for Jev, TypeSafe's System One model, that runs open-weight models on local hardware. It serves the same /v1/systemone endpoint and returns the same typed answers (choice, noul and score) with a probability for every option, so existing Jev clients work against it without changes. A request can also carry images as part of the state, and the questions are then answered about the pictures.
Quoted from the project's README. Read the full README.
Used inJev for Software & DevTools
Highlights
- On 231 public JevBench v1.4 items, gemma-4-26B-A4B with the default options scored 207 correct against 200 for Jev 1.13.
- The same table lists median latency of 93 ms on one laptop GPU, against 665 ms for hosted Jev including the network.
- The README says the two systems disagree on 13 items and the edge is not significant (p = 0.09).
- A choice takes 2 to 62 options. Questions in one request are answered one after another.
- The server does not check API keys and is meant for a trusted network.
Quickstart
bash
docker run --rm --gpus all -p 8765:8765 -v "$PWD/models:/models:ro" \
lichen --model /models/gemma-4-26B_q4_0-it.ggufWatch out
MIT. Needs Docker and an NVIDIA GPU for the published setup. Models keep their own licenses. Jev's score confidence formula is unpublished, so Lichen reuses the choice formula.
Questions about this entry
Short answers drawn from the linked project or page.
The project does not claim significance. On the 231 public benchmark items, the 26B mixture-of-experts Gemma build with the default options scored 207 correct against 200 for Jev 1.13, but the two systems disagree on only 13 items and the README puts p at 0.09. Read it as parity with a much better local latency, not a quality win.
That is the design goal. It serves the same systemone path and returns the same typed answers for choice, yes/no and score with a probability for every option, so existing clients only change the base URL. The catch is trust: the server does not check API keys at all and is meant for a trusted network.
Images are part of the state and each copy of the state shows them before its text. On a hand-checked set of 125 Commons photos it answered all 125, and on 98 randomly drawn road-sign photos it named the right sign 92 times. Add none of these as an option, though, and it missed a quarter of those signs while staying confident about most of the misses.
Two corrections are applied and both were fitted on the project's own 98 questions, not on the benchmark it reports. Answers move toward uniform in proportion to how far the two readings of the question disagree, and label logits are divided by a temperature, 1.25 for the quantized build and 2.25 for the other, with neither changing which answer ranks first. Another model needs its own fit.
Google's gemma-4-26B-A4B on one 24 GB laptop GPU; served from vLLM a decision takes 56 ms at the median, and the same server can hold a 128k-token chat with thinking and images.
Each question becomes one chat prompt listing the answers as labels, and the answer is the probability the model gives each label as its next token, so one forward pass yields a distribution with nothing to parse.
It writes the state and question twice (prompt repetition) and lists every option twice in rotated orders, summing each option's letters, which matches averaging over rotations at half the input tokens.
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 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 Python server in front of llama-server that reads one-token label probabilities and exposes them as POST /v1/systemone.
Jev Models & Reimplementations#community#python#open-models
Community
khimaros/verdictPython
khimaros/verdictPythonA local yes/no decision model that speaks Jev's wire format on a laptop CPU, refusing Choice and Score until those question types ship.
Jev Models & Reimplementations#community#python#open-models
Community
feder-cr/jevC++
feder-cr/jevC++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 Apache-2.0 decision engine with a TypeSafe-compatible endpoint, document answers with source passages, and a ViZDoom demo that picks actions from structured game state.
Jev Models & Reimplementations#community#python#local
Community
deepanwadhwa/OpenDecisionPython
deepanwadhwa/OpenDecisionPythonA self-hosted System One server: one /v1/systemone contract, a Docker image each for Laya and kev, and a playground where snake, dino, and tetris call that endpoint.
Jev Models & Reimplementations#community#python#local
Community
chaitin/DecisPython
chaitin/DecisPythonFrom 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.