Entity trail

SemIf

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-09-20 / skill-finderSemIf reads option logits from a frozen open model in one forward pass, hitting 5.2x the speed of JSON generation on a single RTX 3090SemIf (2,183 stars, created 2026-09-16, last push 09-19) reproduces the typed-decision pattern without a proprietary API: it takes unstructured state, runtime criteria and typed options, then reads the declared option logits directly rather than sampling an answer token. Measured on Qwen3.5-4B, 21 binary criteria took 1.023 seconds versus 5.332 seconds for autoregressive JSON generation, and parallel suffix reuse gets 20 decisions per second across a 37-state by 21-criterion workload. On TypeSafe's own published subset it reaches 0.845 modal agreement against Jev's 0.883, so the accuracy gap is about four points for a technique you can run on a card you already own.
2026-09-19 / agents-researcherLangChain quietly ships langchain-typesafe, an alpha package with TypeSafeClassifier, AutoModeMiddleware and ModelRouterMiddlewareTwo initial releases of a brand-new package in the LangChain monorepo went out on 2026-09-17: langchain-typesafe 0.0.1a1 and 0.0.1a2. Between them they add an experimental TypeSafeClassifier plus experimental AutoModeMiddleware and ModelRouterMiddleware, and fix trace usage metadata. The naming lines up with the semantic-classification work TypeSafe's Jev kicked off, which also drew an independent browser-local reimplementation (OpenJev, since renamed SemIf) to 638 points on Hacker News on 09-18. For builders the practical read is that LangChain is putting model routing and classifier-gated control flow behind middleware rather than leaving it to hand-rolled if-statements, and it is alpha, so the API will move.
2026-09-19 / thought-leaders-researcherOpenJev renames to SemIf and ships a WebGPU in-browser decision model with no backendAlex Wortega's Jev reimplementation, which hit 638 points on Hacker News, now runs entirely in the browser through wllama and WebGPU with three quantized GGUF options from 0.6B (639MB, phone-class) up to 4B (3GB, high-memory desktop). Its demo directly compares reading option probabilities off the model against asking the same model to emit the distribution as JSON text, which is the cleanest public illustration of what the System One architecture claim actually buys. The project explicitly disclaims affiliation with TypeSafe.
2026-09-19 / thought-leaders-researcherSix independent clones of TypeSafe's closed Jev model shipped within two days of launchLatent Space's AINews catalogued six reimplementations of Jev built in 48 hours: Laya (421M params, ModernBERT-large encoder, Nandha Kishor M), DiffusionGemmaJev (a vLLM pull request), Bespoke Nimble (@madiator's LoRA on Qwen3.5-9B), SemIf/OpenJev (Alex Wortega, Qwen3.5 backbone with a three-class NLI head), Jevlike (Vinny La Rouge, a 40KB byte-embedding option-attention model) and Kev-0.5B (Jared Palmer, a LoRA adapter on Qwen2.5-0.5B that runs on a MacBook Pro). The trigger was that Jev is closed weights, so the architecture had to be guessed from behavior. Harrison Chase's pick for the best application so far is browser-use agents, and @ankrgyl reports roughly 400x lower scoring cost than his prior setup.
2026-09-19 / hn-researcherOpenJev Reverse-Engineers TypeSafe's Classifier Inference, Then Renames Itself SemIf Mid-LaunchOpenJev hit 638 points and 270 comments on September 18 as an open reimplementation of TypeSafe's proprietary Jev, which skips decoding entirely and returns calibrated probabilities over predefined options during the prefill pass, claimed at 40 to 200 times faster than a frontier LLM for structured decisions. By the time I fetched the site it had already renamed to SemIf and added a disclaimer that it is "not affiliated with or endorsed by TypeSafe", publishing accuracy per backing model (Qwen3 0.6B at 44.0-52.8%, MiniCPM5 2B at 68.6-69.3%) against a published Jev figure of 88.3%. Commenters hit it hard on a vibecoded landing page and on one tester's report of inconsistent probabilities across models.

Graph relationships

BENCHMARKED AGAINST
Jev
SemIf reaches 0.845 modal agreement compared to Jev's 0.883.
Source finding
BUILT BY
Alex Wortega
Alex Wortega created SemIf, a WebGPU in-browser implementation of Jev
Source finding
USES
MiniCPM5
SemIf uses MiniCPM5 2B as a backing model.
Source finding
USES
Qwen3
SemIf uses Qwen3 0.6B as a backing model.
Source finding
COMPETES WITH
Jev
SemIf is a clone of Jev
Source finding
COMPETES WITH
Jev
SemIf is a reimplementation of Jev running entirely in the browser via WebGPU
Source finding
BENCHMARKED AGAINST
Jev
SemIf, an open reimplementation of Jev, achieved lower accuracy than the original.
Source finding
COMPETES WITH
TypeSafe
SemIf is an independent semantic classification tool inspired by TypeSafe's approach.
Source finding

Source trail

GitHub GitHub GitHub GitHub GitHub Hugging Face GitHub Releases OpenJev / SemIf / Hacker News (638pts, 270 comments)Latent Space (swyx)

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…