Practical overview
When it fits
Use AnyJev when application logic needs structured choices and probability estimates from an LLM, rather than only free-form text.
Who it is for
- Python developers building applications that need language models to return choices from defined options.
- Teams exploring confidence-aware decision handling or routing uncertain decisions to reasoning.
Documented uses
- Represent a model response as a typed choice, yes/no answer or ordered score.
- Retrieve probabilities associated with decision options.
- Batch decisions or use adaptive reasoning for low-confidence Tacit outputs.
Requirements & limits
The project is a public GitHub repository identified as open source under the Apache-2.0 license.
- The documented interface supports choice, yes/no and ordered-score decision types, plus batches.
- Tacit supports Transformers, in-process vLLM and a running vLLM server as serving options.
- Adaptive reasoning can be capped over a recent decision window; the README gives a default cap of 0.2 over 1,000 decisions.
- The documented HTTP gateway listens on 127.0.0.1 and has no authentication of its own.
- The README’s examples focus on Tacit models; they do not provide a concrete setup recipe for wrapping an arbitrary external LLM.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Make a decision with a Tacit model in Python
Load a Tacit model and retrieve an answer, option probabilities, and decision route for a choice.
Before you start
- Python environment with the
anyjevpackage and the modelmorriszjm/Tacit-9Bavailable throughTacit.from_pretrained.
Steps
Import
Tacitfromanyjev.Load the model with
Tacit.from_pretrained("morriszjm/Tacit-9B").Call
decidewith astate, aquestion, and anoptionslist.Read the returned
answer,probs, androutefields.
Expected outcome
A selected option, a probability mapping for the options, and a route value such as `one_forward` or `cot`.
Guide limitations
- The documented example uses the
morriszjm/Tacit-9Bmodel; no broader model compatibility is established here. - The README lists transformers with one GPU for this Python example.
- The README describes adaptive reasoning settings separately; the basic example does not enable adaptive routing.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 13 October 2026
- Original website title
- nokia-applied-research/AnyJev · Official homepage title · retained in its source language