Every Jev guide, in one place
39 guides, grouped by the question each one answers: what Jev is and what a decision costs, how to put it inside an agent, and what people are already using it for. Every one of them is a page on this site, and every figure on them is the one its author published.
Updated · by Jon Kraayenbrink
New here? Read Jev explained, then how to use Jev. Everything else on this page answers one narrower question.
What Jev is, what it returns and what it costs
Start here if you have not made a call yet. Shortest first: the plain explanation, the full one, the name for the pattern, the comparison, the price and the data behind all of it.
Jev explainedThe plain-English version, in three minutes and with no jargon in it.
What is JevThe full explainer: what a call returns, what it costs, how long it takes and where it fails.
What is Jev EngineeringThe name for the split: an LLM writes, Jev decides, code acts.
Jev vs an LLMThe same job done both ways, with the figures each builder published.
Jev pricing$0.042 per million input tokens, output free, and no free tier.- Jev accessWhere to get a key, which routes skip TypeSafe signup, and the signup status by date.
- Jev statisticsThe Build Report: builds counted, cost per decision, latency and stars, recomputed as the directory grows.
- Jev vs OpenAI DecisionsSide-by-side: GA documented vs public beta, confidence scores vs unpublished, text vs vision.
- OpenAI Decisions alternativesWhen to use Jev vs Decisions, and when Decisions still wins on vision.
- Jev alternativesFair comparison: Cloudflare Clef, Perplexity Decisions, OpenAI Decisions, and open-source paths.
- Cloudflare Clef alternativesWho should use Jev instead of Cloudflare Clef, and when Clef wins on vision and open weights.
- Perplexity Decisions alternativesWho should use Jev instead of Perplexity Decisions, and when Perplexity wins on vision.
- Structured Outputs alternativesCategory boundary: chat JSON schema vs decision API, and when each fits.
- Jev vs ClefSide-by-side: images in preview vs audio and video, $0.042/M vs Clef-flash $0.038/M and Clef $0.24/M, closed vs open weights.
- Jev vs PerplexitySide-by-side: both GA decision APIs, ~5% per-token difference, vision support comparison.
- Jev vs Structured OutputsDifferent category: typed questions with probabilities vs arbitrary JSON extraction.
How to build with Jev
From the first call to a harness that runs unattended, then the two questions everyone asks once the bill arrives: can I run it myself, and can I train my own.
How to use JevYour first call line by line, then the SDKs, the gateways and the video walkthroughs.- Jev question typesChoice, Score and Noul side by side, and how to word a question so the answer means what you meant.
- Jev confidenceWhat the confidence number is, why a Noul has none, and where to draw the lines for act, ask and escalate.
Jev with Claude CodeRouting and compaction inside Claude Code, and where the 200× headline stops.- Jev with VercelAI SDK's experimental_decide and AI Gateway's TypeSafe-compatible API, with the builds that run on them.
Jev MCPThe MCP servers that let an agent ask for a typed judgment while it works.
Agent promptsCopy-paste configs for Claude Code, Cursor, Codex, Cline and Roo: skills, MCP and orchestration.
Jev agentic harnessThe decision layer: the model proposes, Jev answers, code acts, with a receipt per step.
Jev orchestrationWho goes next, is this good enough, and when a person has to step in.- Jev routerPicking the model, the tool or the skill before anything is spent, and why routing down and back up can cost more than not routing.
- Jev compactionScoring old tool calls instead of summarising them: how it works, the plugins, and the case against.
Open-source JevThere are no open weights. These are the models that answer typed questions on your own machine.
Fine-tune JevYou cannot fine-tune Jev itself. What people train instead, and what each run cost them.
Jev and imagesYes, in preview: PNG and JPEG behind a header, and the Box type. The limits, and when OCR or a vision model still goes first.- Jev limitationsWhat Jev gets wrong by design — maths, dates, counting, writing — and the fix for each.
What people use Jev for, by job
The same builds the feed carries, read by who you are rather than by what the decision is. Every figure on these pages is the one its author published.
Jev use casesEvery catalogued build, grouped by the job Jev does inside it.
Jev as a judgeJev in place of an LLM judge: evals, rubrics, and what the confidence score is not.- Jev for searchRouting, reranking and RAG passage filtering, with per-query costs and a test it lost.
- Jev for classificationLabels at the per-item cost builders paid, and where to hand off the unsure ones.
Jev for SEOAudits, GEO and slop checks at a per-page cost that makes them routine.
Jev for adsCompetitor libraries read in full, and creative pretested before the spend.
Jev for marketingResearch, testing, publishing, demand and analysis, with the published cost of each.
Sponsored guides: products for building with Jev
Tools, APIs and services from other companies that work with Jev or next to it. Each guide explains what the product does, what it costs, who it fits and where it falls short.
Jev builds, repos and videos, by format
The guides above are ours. These are the lists underneath them: 855 catalogued builds and 181 other people’s guides, docs and talks, filtered by the form they take.
- All buildsThe full feed: every build, guide and post, newest first.
- All Jev builds, by nameEvery one of the 855 builds on one page, by use case, with one line on what each does.
- Jev Engineering reading listEvery guide, talk and build that shaped Jev Engineering, in reading order: the LangChain harness guide, TypeSafe’s founder talks, jev-ultrafast and instant compaction.
- Jev GitHub reposjev-ultrafast, OpenJev, nanojev, jevlike, simplejev, jev-trader and every other repo that calls Jev. Each one linked, with the figures its author published.
- Jev agent skillsAgent skills that build with Jev or hand it the agent’s small decisions. Install TypeSafe’s own with npx skills add typesafe-ai/skills.
- Jev demos and threads on XWatch the Jev demos posted on X without leaving the page. Every card plays the video from the original thread, from browser agents to Doom.
- Jev tutorials and demosEvery Jev tutorial, walkthrough, interview and demo worth watching, collected and playable on this page. Start with the one that matches what you want to build.
- Jev apps you can tryWorking Jev apps, open in one click, no install and no API key. Each one sends Jev a question and shows you the typed answer that comes back.
- Jev docs and cookbooksTypeSafe’s quickstart and docs, the LangChain harness guide, the awesome-jev lists and the launch coverage. What to read before your first Jev call.
Jev use cases, all 21 of them
The same builds again, split by what the decision is for. For the version grouped by the job Jev does rather than by taxonomy, read Jev use cases.
- Agents and browsers181
- Context and memory37
- Coding and code review50
- Routing and model choice68
- SDKs and integrations131
- Open source179
- Benchmarks and evals168
- Search43
- Documents and OCR36
- Security and abuse37
- Inbox and support25
- Social feeds55
- Ads and marketing24
- Sales and leads15
- SEO and GEO10
- Ecommerce7
- Games and real time83
- Trading and markets26
- Robotics and devices15
- UI19
- Tools and apps51
Jev tools you can use here
Not reading: things that run. The tools board lists what other people ship, and 5 free tools run on this site, on Jev, with no key of your own.
- AI Slop DetectorPaste a URL. It screenshots the page, checks 34 tells of AI-generated design and copy, and gives you a slop probability you can share.
- AI Writing CheckerPaste your text. It checks 12 AI writing patterns, quotes each one it finds, and gives you a score for how much it sounds like AI.
- Meta Description CheckerPaste a title and meta description. It measures the pixel width Google cuts at, then Jev judges whether a searcher would click it and what is weak.
- Prompt Injection TesterPaste a prompt, web page or tool result. It finds known injection patterns, quotes each one, and Jev scores how likely the text is an attack on a model that reads it.
- LLM Cost CalculatorEnter calls a day and tokens per call. See the monthly cost on 10 models from vendor list prices, with Jev, which bills input only, in the list.
Made with Jev is independent and not affiliated with TypeSafe AI. Something missing? Submit a build.