Speculative fan-out
Pack many questions — including speculative ones — into one request and let code pick what mattered.
35 rows are filed under this pattern, 35 of them with code. Rows are listed as the catalogue orders them: official first, then rows with code, then by stars.
Filter and search them in the catalogue →
Rows marked “not Jev itself” or “shadow mode” stay listed, but they are not examples of deciding with Jev.
-
Cookbook: Parallel questions
A 13-question regulatory briefing over one long article, showing that batching every question into one call is far cheaper and faster with no change in answers.
-
Pattern: Speculative fan-out
Pack many questions, including ones you may not need, into a single request and let your code decide afterwards what was relevant.
-
Quickstart
The canonical first call: one support ticket, one Choice, one Score and one Noul in a single request, in Python, JS and cURL.
-
AutoGPT TypeSafe blocks
Seven production blocks — choice, score, yes/no, ask-many, route, pick-best, filter — with a UTF-8 byte budget, verbatim wire capture and eleven test files.
-
sub2api: Jev as a moderation endpoint
Drops in as a moderation API by asking many parallel Noul questions in one request, one per hazard category, with an anti-injection prefix on every instruction.
-
jev-ultrafast
A high-speed browser agent from Browser Use: Jev decides the operation and which element to act on, and a small LLM is called only when text must be typed.
-
ai-cookbook: Jev track
A graded course from a first call through each primitive, state shapes and criteria, to ticket triage and a multi-step workflow, mirroring all four official patterns.
-
VideoJev
Open System One video models that score temporal candidates and multiple-choice options for several queries over one shared video prefix, then return intervals, answers, and candidate probabilities without text generation.
-
jev-forge
Not Jev: a toolkit that synthesizes decision data, trains and calibrates small candidate scorers (JevForge-0.8B) and serves them behind a Jev-compatible API, first for picking the next click or form control on a web page. Its benchmark sends the same held-out choice, noul and score questions to Jev.written by a model
-
jev-chat: a tool-calling chatbot with no LLM
A chat bot that does tool calling with no language model anywhere: one request asks the request kind, the tool, and every tool's arguments at once.
-
jev-sift
Classify first. Read selectively. A portable agent plugin and MCP tool for batch text classification.upstream description
-
pi-typesafe
TypeSafe decisions for Pi: batched evaluation tool, terminal playground, and typed API for extension authorsupstream description
-
OneVOneJev
A browser 1v1 FPS where every decision tick judges movement, view angle, aim, fire and jump.
-
system-one
Batched single-token choice inference for open language models, compatible with TypeSafeupstream description
-
slop-grader
Jev-powered, rule-based grader for text files. Runs every rule against every line in parallel. No skimming, no missed lines.upstream description
-
JevGPT: Thinking Fast and Slow in Autonomous Research
Research agent from arXiv 2610.18272. GPT-5.5 plans searches and writes the answer; in BrowseComp-Plus QA, Jev answers a Choice over numbered source lines (where is the answer) and a Noul (does the region address it) per document region and query, in parallel. GPT gets the original lines.
-
duckdb-jev
High-throughput, robust native DuckDB extension for batched and streaming TypeSafe/Jev classification, scoring, and semantic predicates from SQL.upstream description
-
jev-ultralightspeed
Answers one question over a pile of text — tickets, reviews, logs — by packing many items into each request, and reports 32x the throughput of one request per item at 41% lower cost.
-
jev-tree
Recursive Jev choice over a taxonomy. Select from more than 255 options without breaking TypeSafe Jev's choice cap.upstream description
-
jevswiftsdk
Independent, unofficial and dependency-free Swift SDK for the System One API: choice, noul and score questions typed as Swift values, such as a choice over an enum, named batches answered in one call with every answer checked, retries with a policy and cancellation. iOS 15+, macOS 12+ and Linux.written by a model
-
sqlite-jev
Batched natural-language judgments for SQLite, powered by TypeSafe Jevupstream description
-
jackalope
A desktop workspace for coding agents, parallel Git worktrees, and code review.upstream description
-
jev-pr-judge
Typed verdicts on pull requests with TypeSafe System One (Jev): one parallel call, policy in code, usable as a GitHub Actionupstream description
-
snake-jev
Snake controlled by parallel Jev assessments, with one API call per game tick.upstream description
-
jev-switchboard
A JEV-gated semantic communication layer for parallel coding agents.upstream description
-
psearch
Parallel web search for terminals and agents, with local Chromium and Jev-guided exploration.upstream description
-
typesafe-showcase
Next.js UI showing off TypeSafe's System One model (Jev) — parallel Noul judgments and a Choice-based citation checker, deployable to Vercelupstream description
-
A deep dive into Jev, TypeSafe's System One model
The densest independent explainer: code in JS, Python and the AI SDK, all three answer shapes, the advanced patterns, and an honest list of where the model fails.
-
Example: speculative fan-out
Asks for an operation plus a target for each operation it might have picked, so a browser step never needs a second round trip.
-
Example: three primitives in one request
A minimal first call asking a choice, a score and a noul together, annotated with the asymmetries that catch people out.
-
Jev on Cloudflare Workers AI
Workers AI binding and REST samples asking a noul, a choice and a score in one call, with the full response including per-answer confidence.
-
jev-fanout-bench
Measured: asking TypeSafe Jev N questions in one call bills the state once. 2,976 real requests, raw data, exact billing check.upstream description
-
jevcrawl
MCP server and CLI that crawls a website for an agent's question. Per page, one Jev request asks a `noul` for every text block (does it help answer the goal?) and two for every link (leads to the answer, or to a section about it); a last `noul` asks whether the kept excerpts fully answer the goal.written by a model
-
typesafe-image-diffusion
Diffusion-style pixel art out of a general classifier (TypeSafe Jev): 256 parallel pixel questions + refinement passesupstream description
-
Using TypeSafe Jev with the AI SDK
The richest Vercel walkthrough: single and multi-question calls, probability-threshold routing, and unit tests with a mock evaluation model.