User repository on GitHub
Rate and spend limits for code that calls Jev and other decision models
- Forks
- 1
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Repos & SDKs
Decision Gate
Pre-1.0 npm library that every Jev request in a loop goes through: it waits for room under 80% of the account's requests-per-minute and tokens-per-second limits, pauses all local callers on 429 for the server's Retry-After, and refuses requests above a per-key daily spend ceiling of USD 0.20 by default.
- Stars
- 2
- Language
- JavaScript
- License
- APACHE-2.0
- Published by
- Community
- Last verified
- Oct 2026
About this project
Rate and spend limits for code that calls Jev in a loop. If you ask Jev the same kind of question many times, say a go/no-go on every new job posting for each of fifty students, three things go wrong at volume. Requests start failing with 429, and every caller that retries on its own schedule makes it worse. The same posting and question get sent twice, and you pay twice. And nothing stops a runaway loop before the bill does.
Quoted from the project's README. Read the full README.
Works forJev for Decision gating
Used inJev for Software & DevTools
- Languages
- JavaScript
- Platforms
- Vercel AI GatewayAPI
- Capabilities
- Open sourceRuns locallyRequires API key
Highlights
- Defaults: USD 0.20 per day per key, USD 0.02 per run, at most 4 in-flight requests, and requests of 32,000+ tokens go one at a time.
- The 429/529 pause is machine-wide and cross-process: every caller on the account waits the server's Retry-After, or one second when none is given.
- The answer cache stores keyed hashes and numbers, never state or question text; entries last 30 days, and changing scope or model starts fresh.
- Built-in redaction covers known secret formats and emails; every attempt is checked against the never-send list and refused with RedactionError.
- An experimental Vercel AI Gateway provider sends requests through https://ai-gateway.vercel.sh/typesafe/v1/systemone with a separate ledger and rate window.
Watch out
Apache-2.0 and pre-1.0 (0.1.0, so breaking changes raise the minor version); it is unofficial and not affiliated with TypeSafe, and its cache cannot see answers other processes store after it opens.
Questions about this entry
Short answers drawn from the linked project or page.
It waits for room under 80% of the account's RPM and TPS, pauses every caller together on 429 for the server's Retry-After, enforces per-run and per-day USD ceilings, and caches answers so a repeated question over the same state is not paid for twice.
No. The cache stores only keyed hashes and numbers. Requests are checked against a never-send list before anything is sent or cached, and one containing a forbidden value is refused with RedactionError rather than rewritten.
Known secret formats, email addresses, and values after an authorization, proxy-authorization or auth key in two shapes: an HTTP authentication scheme followed by a credential, and a token-shaped run of 16 or more characters. Anything else is kept, so phrases like work authorization: F-1 OPT, OPT-STEM-Extension or a posting's Authorization: must be authorized to work reach the service unchanged.
USD 0.20 per day per key, reset at UTC midnight, and USD 0.02 per run. At the pinned rate of $0.042 per million input tokens with output free, the daily ceiling covers about 4.7 million input tokens and the run ceiling about 476,000, so a long screening job opens a run per batch or raises per_run_usd in the config.
Entries last 30 days, and changing scope or model starts fresh, so rewording any question changes the scope and no old answers are reused. Answers cached through the unpinned Vercel AI Gateway provider expire within a day, because the model behind it can change without notice.
Requests go through https://ai-gateway.vercel.sh/typesafe/v1/systemone with TypeSafe's request and answer shapes, so callers change nothing, and the gateway key is only ever sent there. It has its own ledger, daily ceiling and rate window with defaults of 60 requests a minute and 2 in flight that the README calls conservative guesses rather than measurements, and it sends the floating model id typesafe-ai/jev, so jev.config.pinned is false for it.
Apache-2.0 and pre-1.0 at version 0.1.0, where a breaking change raises the minor version. It is unofficial and not affiliated with TypeSafe, and its cache cannot see answers other processes store after it has already opened.
More Jev resources like this
21GitHub stars
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
chengyongru/fastjevPython4GitHub stars
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
edamame-labs/tab-jevPython2GitHub stars
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
innocentdiaz/s1_rubyRuby1GitHub stars
Dependency-free TypeScript SDK that turns Jev Noul, Choice and Score answers into named decisions: enter/exit thresholds with hysteresis, nested decision trees settled in one call, a JSONL journal, threshold replay without inference and Brier calibration.
Jev Repos & SDKs#community#typescript#decisions
Community
edgardcham/hunchoTypeScript
edgardcham/hunchoTypeScript4.5kGitHub stars
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
crmne/ruby_llmRuby882GitHub stars
A Node and TypeScript package that runs Convai's Laya checkpoint through ONNX Runtime, with the same system_one shape as the Python reference.
Jev Repos & SDKs#community#typescript#open-models
Community
receptron/layaTypeScript
receptron/layaTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
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
From the Jev community
Posts from builders shipping with Jev right now.