How does jev-bisect choose each next number to test?

After a higher or lower answer the current guess is excluded and the next guess is the floor midpoint of the remaining candidates: for min=0 and max=100 the choices evaluate 50, 75 and 62. Endpoints on the grid stay reachable.

Can I pin a Jev version or reuse a TypeSafe client?

Yes. Pass model='jev-1.13.0' to override the default jev-latest, and optionally supply a TypeSafeClient context so one connection is reused across searches; a supplied client remains open after the call.

What happens if Jev is wrong, or the bounds are too wide?

Jev can make incorrect comparisons, so a wrong answer surfaces as MaxTurnsExceededError or SearchExhaustedError. Bounds are never expanded by the search, so choose a range that contains the rounded answer, and a large range or fine precision can need more guesses than max_turns allows.

How do I search decimals, and what does precision actually mean?

Candidates are multiples of precision measured from zero: 1 searches whole numbers, 0.01 searches hundredths, 2.5 searches 0, 2.5 and 5. Jev is told to round to the nearest multiple with halfway values going away from zero. Precision is search resolution, not an error tolerance, and bounds snap inward to the grid.

How do I install it, and what comes back at the end?

Python 3.10 or newer, pip install jev-bisect, and TYPESAFE_API_KEY exported. min and max are required and must satisfy min < max. A result returns only after Jev selects exactly, carrying answer, turns, the final state and a history where each entry records state, choice, confidence, probabilities and model version.

Can I drive the search myself without spending an API call?

Yes. advance(choice, min=..., max=...) applies one step locally and returns a SearchState, defaulting to the bounds' midpoint when last_guess is omitted. When the returned bounds are equal, pass last_guess to evaluate the last candidate: exactly returns that state, anything else raises SearchExhaustedError.

What license is it under, and can I run its tests offline?

MIT. The tests use deterministic responses and a mock HTTP transport, so they need no API key and make no network requests. SDK retries are disabled for these calls, so one evaluated guess is one turn, including the initial midpoint.

CommunityJev Repos & SDKs

Jev Bisect

A Python library that turns Jev into a bisection oracle: it repeatedly asks higher, lower or exactly about the midpoint of a numeric range and returns the answer, turn count and full decision history. Built on the official TypeSafe Python SDK and installable from PyPI.

See highlights

Open github.com

Stars: 0
Language: Python
License: MIT
Published by: Community
Last verified: Oct 2026

mkly/jev-bisect

User repository on GitHub

View on GitHub

A Python library that uses Jev to find a numeric answer through repeated higher, lower, or exactly decisions.

Forks: 0
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

Find numeric answers with Jev through repeated higher, lower, or exactly decisions. Supply a question and bounds; the library calculates the guesses using the official TypeSafe Python SDK. Requires Python 3.10 or newer. Install from PyPI: min and max are required keyword arguments and must be finite numbers with min < max. Omit lastguess or pass None to start at the midpoint. To choose the first guess, supply an override such as bisect(question, min=0, max=100, lastguess=25).

Quoted from the project's README. Read the full README.

Tagscommunity bisection numeric-search

Works forJev for Classification

Used inJev for Software & DevTools

Languages: Python
Platforms: API
Capabilities: Open sourceRequires API key

Highlights

Each guess is one typed higher/lower/exactly decision; a result is returned only once Jev selects exactly.
Install with `pip install jev-bisect` and set TYPESAFE_API_KEY; guesses use exact rational arithmetic and integer candidate indexes internally.
Configurable precision (default 1) and max_turns (default 20); each evaluated guess counts as one turn and SDK retries are disabled.
Returns answer, turns, final state and history entries with state, choice, confidence, probabilities and model version.
Tests use deterministic mock responses and a fake HTTP transport, needing no API key or network.

Watch out

MIT-licensed and needs your own TYPESAFE_API_KEY. Bounds are not expanded by the search and Jev can make incorrect comparisons, surfacing as MaxTurnsExceededError or SearchExhaustedError.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

RubyLLM

4.5k

Ruby-native AI framework whose RubyLLM::Judge class asks typed judgment questions: a probability that a statement holds, choices or scores, answered by the TypeSafe provider alongside 18 other hosted and local providers under one consistent API.

Jev Repos & SDKs#community#ruby#framework

Community

crmne/ruby_llmRuby

FastJev

Self-hosted Python SDK and System One-compatible API for runtime-defined Choice, Boolean and Score decisions on pinned open models, across Torch, vLLM, MLX, llama.cpp and WebGPU. Each result records probabilities, token usage, timing, model revision and prompt version, with committed row-level benchmarks.

Jev Repos & SDKs#community#sdk#self-hosted

Community

chengyongru/fastjevPython

tab-jev

A pre-release Python framework for predictions on mixed text and tabular data: Jev Noul and Score questions turn text columns into numeric features, then a tabular foundation model such as TabPFN learns from labeled rows in context. Offers hosted and fully local backends.

Jev Repos & SDKs#community#tabular#predictions

Community

edamame-labs/tab-jevPython

S1 Ruby

Ruby gem that makes System One measurement a language primitive: psi prepares a state, judge/choose/score return calibrated distributions, and ? collapses them for control flow. Ships TypeSafe, protocol (/v1/systemone) and CUA providers behind one interface with its own spec suite.

Jev Repos & SDKs#community#ruby-gem#sdk

Community

innocentdiaz/s1_rubyRuby

CloJev

Unofficial portable Clojure SDK for TypeSafe System One. It posts state and typed questions to /v1/systemone, lists available models, and returns validated Clojure data with keyword or string answer keys, on a JVM backend or a native Jolt backend.

Jev Repos & SDKs#community#clojure#sdk

Community

antlobach/clojevClojure

jev-sdk-go

Dependency-free Go 1.24+ client for Jev Noul, Choice, and Score questions: ChoiceOf[T] and ScoreOf[L] return the caller's own types and reject labels the question never offered, with validation before sending, retries that honor Retry-After, OpenRouter support, and eleven tested examples.

Jev Repos & SDKs#community#go#sdk

Community

HomayoonAlimohammadi/jev-sdk-goGo

From the Jev guides

Original write-ups on what this entry is for.

The Open Reimplementations: Local Models That Speak System One

Sep 22, 2026

Closed weights have not stopped the ecosystem from rebuilding the shape in the open. API-compatible servers, retrained heads on small models, and inference-time tricks that skip generation entirely.

Jev Repos & SDKs#open-models#research#local

Read article

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai