ardana
Ollama-style single binary that pulls and runs open decision models locally, with a TypeSafe-compatible API and playground.
Overview
Ardana runs System 1 decision models on your own machine the way Ollama runs chat models. You pull a model by name, give it text and a few questions, and it reads everything once and answers each question with a probability for every option. It never writes text, so there is nothing to parse or retry.
Questions are choice (2 to 255 named options), score (2 to 10 ordered levels, with a score along the scale) or noul (probability of yes). The CLI prints confidence and certainty alongside probabilities. The README says the first run downloads a 1.3 GB model, after which an answer takes about a tenth of a second on an M1 MacBook. Models include a decider family and Qwen 3.5 in several sizes and quantizations. The server's API is compatible with TypeSafe clients, so you can point the TypeSafe SDK at it.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which department should handle this? | billing · technical · sales |
| Yes / no | Does the customer ask for a refund? | yes · no |
How it works
- Install the binary and pull a model
- Run a question from the CLI or start the server
- Model answers all questions in one pass
- Use probabilities in code via a TypeSafe-compatible client
Strengths
- Fully local, one binary
- About 0.1 s per answer on an M1 per the README
- Works with the TypeSafe SDK by changing the base URL
Caveats
- Local model quality differs from hosted Jev
- First run downloads 1.3 GB
Getting started
curl -fsSL https://ardana.ai/install.sh | sh, then ardana run decider "<text>" --choice "<question>" opt1 opt2 --noul "<question>"At a glance
- Type
- Service
- Status
- beta
- Runtime
- CLI + local server
- Has code
- yes
- License
- Apache-2.0
Use cases and integrations
How curated lists describe it
Run System 1 decision models locally. Pull a model by name and get calibrated probabilities for your questions in one forward pass. One binary with a TypeSafe-compatible API and a playground.
hellogumbo/awesome-jev
Frequently asked questions
Is it like Ollama?
Yes, for decision models: pull by name and run locally.
Can I use the TypeSafe SDK?
Yes, set TYPESAFE_BASE_URL to your local server.
How fast is it?
About a tenth of a second per answer on an M1 MacBook, per the README.
Related projects
llama.cpp-based C++ server that turns any LLM into a System One decision model, reading answers from next-token logits.
ProjectbetaC++MLX library that turns an existing local model into a decision engine by reading candidate logits, with no fine-tuning, on Apple Silicon.
ProjectprototypePythonLocal server (xks) implementing Jev's /v1/systemone contract on an open-weight LLM, accelerated with Xeon Phi cards.
ProjectresearchRustEducational C11 library and CLI that turns DiffusionGemma 26B into a CPU-first OpenJev-style decision engine reading candidate logits.
ProjectresearchCExperimental engine that makes an existing LLM do Jev-like classification and scoring by reusing context and building JSON.
ProjectresearchPythonReads typed decisions from one forward pass of PrismML's 1.75-bit Bonsai 27B under llama.cpp, using the OpenJev/SemIf logit method.
ProjectresearchPython