01 / Understand the model
Introducing System One Models and Jev
Read TypeSafe's launch material for the typed-decision interface and the official claims.
officialORIENTATION
Start here: official and explainers
The launch material, the docs and the best explainers of the typed-decision contract.
A SHORT READING PATH
Three records to get oriented
Start with the source, inspect a working integration, then read a result with its evidence label attached.
02 / See an integration
jev-use for computer use
Follow one public repository to see where Jev sits in a computer-use workflow.
source-backed
03 / Read a reported result
8,790 SEO decisions
See how a builder's measured workload and its author-demo evidence label appear together.
author demoThen explore benchmarks and independent tests or how Jevland labels evidence.
From this task to a first run
Choose a project, read its requirements, then follow its setup reference or a related guide.
- @typesafe-ai/sdk (TypeScript): How to run →
Node.js 20+ and npm
- typesafe-sdk (Python): How to run →
Python 3.10+ and uv
- Jev Search: How to run →
Node.js 22.12+, pnpm 10.8.0 and a Cloudflare login
- PlayJev: a 0.8B visual decision model for ten games: How to run →
A modern browser for recorded playback
- minojev: a trained decision head on a frozen language model: How to run →
A modern browser for the domain playground
- MemSearch: Jev memory reranking: How to run →
Python environment with pip; a folder of Markdown notes
- Agent Squad: typed Jev routing: How to run →
An existing TypeScript/Node project, or Python environment
- Oko: local code search and prompt-time retrieval: How to run →
Supported macOS or Linux binary; a local source checkout
- Fastbrowse: browser tasks with typed decisions and citations: How to run →
uv; Python 3.13+ fetched by uv; cloud browser or local Chrome
- GPT Researcher: Jev context selection: How to run →
Python 3.12+ and pip
Try a walkthrough
- Connect Jev to your app →
Set up the Python SDK, make a typed routing decision and keep the result under your application's control.
- Select relevant context with Jev →
Score a small shortlist of passages, preserve their source IDs and pass selected evidence to your answering model.
- Connect Jev with JavaScript and TypeScript →
Make a server-side structured decision with the official SDK, then check the response before your app uses it.
- Route the next step of an agent →
Let Jev choose a proposed next step, then apply an explicit allow, review or fallback policy before any tool runs.
Launch post, docs, evals, SDKs and the first explainers — the shortest path into the ecosystem.
44 entries
officialofficial
Introducing System One Models and Jev
The launch source for Jev, its typed decision interface, pricing and TypeSafe workflow evaluations.
officialofficial
Jev documentation
Current product documentation for state, typed questions and the API surface.
postofficial
Jev launch announcement
The launch post and primary product video from TypeSafe's co-founder.
articlesource-backed
A practical Jev walkthrough
Independent walkthrough of the model contract and early experiments.
articlesource-backed
Jev clearly explained
A detailed guide to state, Choice, Score, Noul, probabilities, routing, confidence gates and shadow rollout.
videosource-backed
What is a decision model?
A platform explainer contrasts confidence-scored yes/no and multiple-choice decisions with free-text generation.
videoauthor demo
Jev tutorial: API to voice browser
A long tutorial covering the API, a voice browser, AI memory and a YouTube predictor.
officialofficial
TypeSafe emerges from stealth with $40M
The launch-day press release: a $40M seed round led by DCVC, the Jev launch, and the three co-founders (Diogo Almeida, Erik Gafni, Sasha Sheng).
officialofficial
The Bitterest Lesson
A pre-launch essay from the company blog that extends Sutton's bitter lesson into a hierarchy: right task, then data, then compute, then algorithms, using InstructGPT versus GPT-3 as the example.
benchmarkofficial
TypeSafe workflow evals dashboard
The public dashboard behind the launch numbers: four workflows (security incidents, agent-trace observability, invoice processing, customer service) and ten model configurations compared on accuracy, cost and latency.
repoofficial
typesafe-sdk (Python)
The official Python client (`uv add typesafe-sdk`) with sync and async variants; the quickstart classifies support tickets into billing, technical or other.
repoofficial
@typesafe-ai/sdk (TypeScript)
The official TypeScript/JavaScript client, shipping ESM, CJS and type declarations for Node 20+; same ticket-classification quickstart as the Python SDK.
articlesource-backed
How to use Jev: a practical guide
The most complete third-party tutorial found: installing the Python and TypeScript SDKs, calling `system_one` with Choice, Score and Noul, and a set of launch-week examples.
articlecommentary
TechCrunch: a new kind of AI model is thrilling developers
A developer-reaction piece with named quotes, including Vercel's safety-classifier comparison and other early adopters.
articlecommentary
The Register: TypeSafe debuts a model for machines that plays Doom
Launch coverage that correctly notes the Doom demo runs on structured game state rather than pixels and reports the funding round.
postcommentary
Hacker News launch thread
The launch discussion (1,889 points, 495 comments at fetch): whether "can't hallucinate" is just type safety, whether this is constrained decoding, and what the Doom demo actually shows.
articlecommentary
Latent.Space AINews: a System One model that only decides
The newsletter's lead item frames Jev as a cheap, calibrated inference engine for structured choices, complementary to System 2 LLMs.
articlecommentary
Jev, sorted: what is real and what is still a claim
Argues the headline multipliers are real but chosen against the slowest and priciest baselines; against TypeSafe's own intelligence peer the gains shrink.
repoofficial
Official agent skills repository
TypeSafe's public skills repository with the `typesafe-ai` SKILL.md used by Claude Code, Codex and other agent environments to design bounded Jev decisions.
articlecommentary
DataCamp: the System One model that never hallucinates
A long explainer with a comparison table and the API shape, written from public material.
articlecommentary
Forbes: the $200M startup fixing AI overconfidence
The only source found with a valuation: a $40M seed at about $200M, DCVC lead, plus founder interview details (four years at OpenAI, company founded in 2024).
videocommentary
Gary Explains: Jev from TypeSafe
An explainer video from a well-known hardware and software explainer channel; the title signals a caveat framing.
officialofficial
Jev 1.13 known weaknesses ("jaggedness")
An official docs page listing failure modes of jev-1.13 with workarounds: literal reading of implied conditions and negations, unreliable arithmetic and counting, dates treated as text, weak multi-hop indirection, degradation with irrelevant state, prompt-injection vulnerability, and inconsistent invariants between related questions.
officialofficial
TypeSafe status page
Uptime and incident history for api.typesafe.ai and console.typesafe.ai.
articlesource-backed
Vercel: what is Jev?
Vercel's explainer frames Jev around Kahneman's System One, shows how to structure bounded decisions and separate evidence from questions, and points to the AI SDK evaluate API and the `@ai-sdk/typesafe-ai` provider.
articlecommentary
MarkTechPost: TypeSafe AI releases Jev
A launch write-up covering the three primitives, RLCD training, pricing and latency, pointing to jev-ultrafast and mobile-jev as early adopters.
articlecommentary
ThursdAI: a ChatGPT moment for decisions
A newsletter and podcast recap treating Jev as the lead story and summarising use cases seen in the first 48 hours: computer use, email classification, context compression, safety filtering.
articlecommentary
Warmer Sun: typed decisions, not chat
A primer with an essay and research appendix: API shape, launch pricing, founder background, co-founders and the funding round.
articlecommentary
Jev Engineering: a 10-step setup roadmap
An X Article walking from waitlist and API key through the official skill, the Python SDK, confidence thresholds and the three primitives (choice, score, noul) to a first agent loop.
postcommentary
How to use Jev, and where the 100x actually is
Ten-minute setup (waitlist, official skill, API key, 'use the TypeSafe skill' in the prompt), then the argument: the gain comes from deleting LLM calls that only pick something — next tool, is-this-spam, is-this-chunk-relevant, needs-a-human, is-this-diff-risky — and replacing each with a typed question.
articlecommentary
How to master Jev (full guide)
A long-form X Article covering what Jev is, how the typed-question API differs from prompting an LLM, and where to slot it into existing agents.
integrationofficial
Vercel AI Gateway: Jev free until 25 September
Vercel's announcement that Jev runs at no cost on the AI Gateway until 25 September, calling it the fastest-adopted model on the Gateway.
videocommentary
Diogo Almeida's 17-minute talk: after the assistance era
The founder's 17-minute talk, re-shared with commentary: why RLHF-era models carry a human-shaped flaw, why coding assistants are still the assistance era, and the x200 faster / x400 cheaper / zero-hallucination framing for what comes next.
articlecommentary
You could have built Jev
A technical write-up arguing the Jev recipe was within reach of any team — the piece Eric Zhang says explained the thing he had built the day before.
articlecommentary
Akita on Rails: 'it's a classifier'
The counter-take from a well-known Brazilian developer: Jev works fine, but it is a classifier, and the revolutionary framing is marketing.
postcommentary
LLMs vs Jev, clearly explained
Jev does not generate text: it evaluates predefined decisions directly and in parallel. A failed-deployment example (urgent? which team? risky command? done?) shows Choice, Score and Noul, and why the probabilities matter as much as the answers — 91/9 routes automatically, 52/48 escalates.
articlecommentary
Jev use cases, clearly explained
An X Article walking through the main Jev use-case families with examples. 160k views.
postcommentary
The ten steps of the Jev Engineering roadmap, summarised
A ten-point digest of codila's roadmap: LLMs create, agents act, Jev decides; three primitives; swap Jev in without rebuilding the graph; batch decisions (13 questions 10x faster and 12.2x cheaper in one test); bounded forks; benchmark the loop; rank wide, read narrow; keep Jev out of math, writing and irreversible execution.
integrationsource-backed
System One for TypeScript: provider-neutral decisions
A TypeScript System One runtime provides shared Choice, Noul and Score builders for Jev, Reflex and compatible endpoints, with a companion Pi extension.
integrationsource-backed
Stumble jev-go: Go SDK and interactive CLI
A community Go SDK implements Jev questions through TypeSafe and Vercel transports and includes an interactive command-line wizard.
integrationsource-backed
go-jev: Go client and UNIX decision pipelines
A Go SDK pairs typed Jev requests with a CLI for line-by-line classification, semantic grep and JSON decision output in shell pipelines.
articlesource-backed
Evaluating Jev across 37 public datasets
An independent evaluation of pinned jev-1.13.0 across classification, routing, inference, moderation, legal clauses and rubric scoring. The paper uses frozen task templates, reports calibration and selective prediction, and releases its harness and raw model responses.
articlesource-backed
Jev in the Wild: a study of the emerging project ecosystem
A research paper studying how public GitHub projects use Jev as a reusable decision component. It categorizes applications, typed interfaces and decision purposes, then compares project distribution with public attention.
articlesource-backed
NumericJev: numerical decoding through structured range choices
A training-free algorithm that recovers numerical outputs by recursively narrowing a range through a multiway tree of Jev-like choices. The paper preserves the original question in context and studies how range selection differs from direct candidate selection.