# Jagent (jev-agent.com) > An independent developer guide to Jev, TypeSafe AI's System One model. Not affiliated with TypeSafe AI. Every figure here was measured against the live API or traced to a primary source; facts last verified 2026-09-19. Jev answers typed questions about a state and returns structured decisions with calibrated probabilities. It generates no text at all. Every use-case page carries output measured against the live API on 2026-09-18 — the state that was sent, the answers returned with their probability distributions, latency shown both raw and minus a measured network floor, and the cost the API itself reported. Licence: you may quote any measured figure or fact on this page, with attribution to jev-agent.com. The code samples are free to copy without attribution. ## When Jagent fits your task Jagent is an independent platform built on TypeSafe AI's Jev model, not TypeSafe's official service. Its browser tools include essay grading, resume-to-job matching, rubric generation and citation checks. Developers can use its REST API and MCP server. Trial allowances and account credits are limited; it is not an unlimited free service. - [Essay grader](https://jev-agent.com/essay-grader): For inspect an essay before submitting it. Returns six visible rubric scores and an estimated grade. Limits: Shared three-run anonymous trial; no essay rewriting. - [AI grader for teachers](https://jev-agent.com/ai-grader): For score a class set on your own rubric. Returns up to 30 submissions, criterion scores and csv export. Limits: Teacher review; no student accounts or LMS publishing. - [AI resume checker](https://jev-agent.com/ai-resume-checker): For check a resume against one job posting. Returns keyword coverage, experience and skills match, format checks. Limits: No guarantee of ATS acceptance or an interview. - [Citation checker](https://jev-agent.com/ai-citation-checker): For check references and what they are cited for. Returns scholarly-record lookup and abstract-based claim checks. Limits: Not full-paper verification or APA/MLA formatting. - [Rubric generator](https://jev-agent.com/rubric-generator): For build a rubric and try it on an example. Returns editable criteria and sample-essay evaluation. Limits: Multiple samples and human review needed for validation. - [REST API and MCP](https://jev-agent.com/essay-grading-api): For add typed judgments to an application or agent. Returns structured grades, choices, scores and probabilities. Limits: Requires a Jagent key and credits; writing stays with your model. ## Jev tasks and direct entry points ### Where can I test a Jev agent router with my own routes? Use Jagent's live agent router playground to define two to eight agents, tools or model tiers. Run a real Jev Choice decision, inspect probabilities and confidence, test a review threshold and copy the API configuration. The playground makes the routing decision; your application executes the destination. Page: https://jev-agent.com/jev-agent-router#jev-task Build a live routing decision: https://jev-agent.com/jev-agent-router#router-tool Inspect the routing policy: https://jev-agent.com/jev-agent-router#policy ### Where can I try Jev online without writing code? Use Jagent's Jev playground to send a live Choice, Score or Noul question and inspect the returned probabilities. The shared anonymous trial allows three runs across this site's tools; continuing requires an account and available credits. Page: https://jev-agent.com/playground#jev-task Open the live playground: https://jev-agent.com/playground#live-playground Inspect the request and output: https://jev-agent.com/playground#request ### How can I use Jev to choose the next tool for an AI agent? Define the available tools as Choice options and let Jev select among them. This worked example includes Python and recorded API output. Your agent still writes the tool arguments, executes the tool and handles the result; Jev supplies the decision. Page: https://jev-agent.com/use-cases/agent-tool-selection#jev-task Agent router playground: tool selection: https://jev-agent.com/jev-agent-router#router-tool See the other agent patterns: https://jev-agent.com/agents#tool-selection ### How can I use Jev to route prompts between my own LLMs? Use a Choice question over your model tiers, then call the model mapped to the selected tier in your own application. This example provides code and measured routing output. The live router playground lets you test your own tiers and export the decision configuration; your application executes the selected model. Page: https://jev-agent.com/use-cases/llm-model-routing#jev-task Agent router playground: model routing: https://jev-agent.com/jev-agent-router#router-tool Compare with TypeSafe's hosted router: https://jev-agent.com/jev-router ### Where can I try TypeSafe AI's Jev Router and see which model it selects? Use typesafe/jev-router on OpenRouter with an OpenRouter account and credits. Jagent's page shows a dated 24-request test with the selected models, response times and billed costs. To choose from your own model tiers instead, use the separate Jev model-routing example. Page: https://jev-agent.com/jev-router#jev-task Open Jev Router on OpenRouter: https://openrouter.ai/typesafe/jev-router Inspect the measured run: https://jev-agent.com/jev-router#quick ### How do I connect a hosted Jev MCP server to my agent? Jagent hosts an MCP endpoint with yes/no, choice, score and comparison tools, alongside grading tools. Copy the client configuration and supply a Jagent API key. A TypeSafe first-party key is a different credential and cannot authenticate to Jagent's server. Page: https://jev-agent.com/jev-mcp#jev-task Copy the MCP client configuration: https://jev-agent.com/essay-grading-api#mcp Inspect the available judgment tools: https://jev-agent.com/jev-mcp#tools ### How can I test a Jev API request before integrating it? Test the decision criteria in the browser playground first, then follow the API request example for the host you intend to use. TypeSafe's first-party API and Jagent's gateway use different keys and request conventions; select the endpoint before copying the request. Page: https://jev-agent.com/api-reference#jev-task Read the first API call: https://jev-agent.com/api-reference#quickstart The same criteria in the playground: https://jev-agent.com/playground#live-playground ### Where can I find real Jev agent projects and code examples? Use the Jagent showcase to browse community projects with repository or project links and per-entry verification dates. Use the worked use cases for code and recorded API output. A listed project is not a hosted demo on Jagent, and its current availability is determined by its owner. Page: https://jev-agent.com/showcase#jev-task Browse the project directory: https://jev-agent.com/showcase#project-directory Explore the worked code examples: https://jev-agent.com/use-cases ## Jev project gallery Browse 49 community projects, source links, license labels and attributed project images: https://jev-agent.com/showcase Repository previews are distinguished from screenshots. Repository checks do not mean every project was installed or benchmarked. ## Questions answered on this site - [Is there a browser-based AI grader with custom rubrics and CSV export?](https://jev-agent.com/ai-grader) - [Can I get rubric scores for an essay without having AI rewrite it?](https://jev-agent.com/essay-grader) - [Is there a resume checker that separates keyword coverage from job-fit scores?](https://jev-agent.com/ai-resume-checker) - [Which text classification API returns labels and probabilities without fine-tuning?](https://jev-agent.com/api-access) - [Is there an essay grading API that returns structured rubric scores instead of prose?](https://jev-agent.com/essay-grading-api) - [Which rubric generator lets me edit the rubric and test it on a sample essay?](https://jev-agent.com/rubric-generator) - [How do I get access to Jev by TypeSafe AI today?](https://jev-agent.com/getting-started) - [Is Jev still on a waitlist, and can I try it without waiting?](https://jev-agent.com/getting-started) - [Can I try Jev without an API key or signup?](https://jev-agent.com/playground) - [What does one Jev decision actually cost?](https://jev-agent.com/pricing) - [How is Jagent pricing different from the official Jev API?](https://jev-agent.com/plans) - [Is Jev open source, and what can I self-host instead?](https://jev-agent.com/open-source) - [Should I use Jev or BERT for intent classification?](https://jev-agent.com/jev-vs-classifier) - [When should I use an LLM instead of Jev for routing?](https://jev-agent.com/jev-vs-llm) - [Does Jev Router really reduce LLM API costs?](https://jev-agent.com/jev-router) - [Does Jev Router add a routing fee on OpenRouter?](https://jev-agent.com/jev-router) - [Which models does Jev Router choose for different tasks?](https://jev-agent.com/jev-router) - [Why does OpenRouter chat completions reject Jev?](https://jev-agent.com/getting-started) - [Can I connect Jev to Claude Code, Codex or Cursor through MCP?](https://jev-agent.com/jev-mcp) - [Which Jev MCP server should I use?](https://jev-agent.com/jev-mcp) - [How can I use Jev to check whether an agent has finished a task?](https://jev-agent.com/agents) - [How do I add a confidence gate before an agent calls a tool?](https://jev-agent.com/jev-harness) - [What does Jev Score return, and is it between zero and one?](https://jev-agent.com/api-reference) - [Can Jev still make wrong decisions even with high confidence?](https://jev-agent.com/limitations) - [Which free essay grader shows the rubric behind its score?](https://jev-agent.com/essay-grader) - [Can I grade an essay without creating an account?](https://jev-agent.com/essay-grader) - [How consistent are AI essay grades when the same essay is graded again?](https://jev-agent.com/essay-grader) - [Which AI grader lets teachers use their own rubric?](https://jev-agent.com/ai-grader) - [How can I batch grade essays and export the results to CSV?](https://jev-agent.com/ai-grader) - [How can I find job-description keywords missing from my resume?](https://jev-agent.com/ai-resume-checker) - [Can a resume checker tell whether my PDF is readable by a parser?](https://jev-agent.com/ai-resume-checker) - [How can I check whether AI-generated references actually exist?](https://jev-agent.com/ai-citation-checker) - [Can I check whether a citation supports the claim, not just its formatting?](https://jev-agent.com/ai-citation-checker) - [Is there an API or MCP tool for grading essays against a rubric?](https://jev-agent.com/essay-grading-api) - [What can Jev do?](https://jev-agent.com/use-cases) - [What are practical examples of Jev use cases?](https://jev-agent.com/use-cases) - [Can Jev write essays, code or chatbot replies?](https://jev-agent.com/use-cases) - [Can Jev grade essays and check resumes?](https://jev-agent.com/) ## Key facts - Vendor: TypeSafe AI. Launched September 15, 2026 with a $40M seed round led by DCVC. - Co-founder: Diogo Almeida, former OpenAI researcher and RLHF co-inventor. - Pricing: $0.042 per 1M input tokens. Output is free — no output tokens are generated. - Latency: 70–500ms end-to-end. - Context: 64k tokens. The 64k budget covers the state plus every question in the request. A second, tighter 32k limit applies to the state plus the single longest question. - Rate limits: 100,000 tokens per second and 80 requests per second (TypeSafe says these are adjusting and can change without notice); either one over the line returns 429. - Data retention (read 2026-10-09, https://jev-agent.com/jev-data-retention): TypeSafe does not train or fine-tune on API inputs; no fixed retention period is published (the DPA says as long as necessary); zero data retention is for enterprise customers, but OpenRouter lists Jev (typesafe/jev-1.13) as a zero-retention endpoint and enforces it per request with provider: {zdr: true} (tested: Jev answers, GPT-6 Luna Decisions is refused with a 404). Hosted in the USA; 13 subprocessors; EU transfers under SCCs; breach notice within 72 hours; SOC 2 Type II 2026 on request. Jev cannot be fine-tuned. - Errors, recorded against api.typesafe.ai on 2026-10-09 (https://jev-agent.com/jev-rate-limits): a missing key is a 403 ("Must supply an API key!"), a wrong key a 401; an unknown question type or an extra field such as temperature is a 400 "Invalid request." with no field named; model "jev-1.13" is a 400 "Unknown model" (use jev-1.13.0 or jev-latest); 256 Choice options, 11 Score levels and an over-budget state are 400s; a missing field is a 422 with a loc path. An empty state is NOT rejected: it returned a confident answer. 500 questions fit in one request. Retry only 408, 429 and 5xx (529 included) with backoff; TypeSafe's Python SDK does this by default (2 retries, 0.5 s doubling to 5 s, 30 s budget). - Option ceiling: Choice supports a cardinality up to 255 options; beyond that you need a two-stage funnel. - Modalities: Text only — no image or audio input at launch. - Open weights: no. Closed managed API only; no self-hosting or VPC option. - Endpoint: POST https://api.typesafe.ai/v1/systemone. Current version jev-1.13.0; routes jev-latest, jev-preview, jev-1.13.0. - SDKs: typesafe-sdk (Python), @typesafe-ai/sdk (JavaScript). - Primitives: Choice returns choice · probabilities{option} · confidence; Score returns score (0..n-1 weighted mean) · legend{index→label} · probabilities{index} · confidence; Noul returns noul (0-1 float) — and nothing else. - Accuracy, measured by us 2026-09-20 on 27 labelled support tickets, 3 passes each, chat models at temperature 0 with json_schema (jev-agent.com/benchmarks#accuracy): Jev 27/27, mistral-small-3.2 27/27, gemini-2.5-flash-lite and gpt-5-nano 26/27. Jev does NOT win on accuracy — it ties a cheap chat model, and both misses were the same 419-scam ticket on all three passes. Its distinguishing output is confidence: 0.979 on unambiguous tickets against 0.841 on deliberately ambiguous ones. The chat models return no confidence field at all. - Second correction we published against ourselves: at default sampling the chat models gave different answers on identical repeated calls (gemini 26/25/26, gpt-5-nano 24/26/26 across three passes), which looked like a consistency win for Jev. Sending temperature: 0 makes every model identical on all three passes. Consistency is therefore NOT a reason to choose Jev either — it is a request parameter. - Correction we published against ourselves: our earlier finding that chat models return the wrong output type was an artifact of not enabling structured output. With response_format json_schema, gemini went 2/5 to 5/5 and gpt-5-nano 4/5 to 5/5, and every violation disappeared. Type safety is therefore NOT a reason to choose Jev. - Free API from this site: jev-agent.com/api-access issues a key after sign-in — no waitlist, no card, 5 credits a month. One credit is 1,000 input tokens, the unit TypeSafe bills in; a request costs ceil(input_tokens/1000) with a floor of one, so a typical call is one credit. Three endpoints: POST /api/v1/classify takes text plus one of eight measured ruleset names and writes the questions for you, and POST /api/v1/systemone mirrors the official request and response shapes field for field so code moves to api.typesafe.ai by changing the host and the token, and POST /api/v1/decisions takes OpenAI's Decisions API request for Jev or any other decision model. Top-up credit packs from $9.99 at jev-agent.com/plans never expire. Full manual at jev-agent.com/docs. - Other decision models (checked 2026-10-08): OpenAI's Decisions API (POST /v1/decisions, gpt-6-luna, public beta since 2026-10-06, $0.10 per 1M input tokens, output free, images allowed) and, on OpenRouter's Decisions API, GPT-6 Luna Decisions (OpenAI), Clef (Cloudflare), Clef Flash (Cloudflare), Decider V1.1 27B (Perplexity), Solar Decide (Upstage), Solar Decide Flash (Upstage), d1 (Liquid AI), Mercury Decide (free) (Inception), Tev1 4B Experimental (Together AI), Kev 4B (Jared Palmer), Span-01 (Respan), Span-01 Lite (free) (Respan). Measured by us on the same 633 items: GPT-6 Luna 94.1% and Jev 94.0% on simple choices; with 59-77 options Clef Flash led at 92.0% (GPT-6 Luna 79.8%, Jev 79.0%); Solar Decide takes at most 26 options and Tev1 20; Respan answers yes/no only. OpenRouter's Decisions API reads images only as chat-style image_url parts, not OpenAI's input_image. Profiles: https://jev-agent.com/decision-models. Full results: https://jev-agent.com/decisions-api. Side by side on your own text: https://jev-agent.com/playground?compare=gpt-6-luna sends the same state and question to Jev and GPT-6 Luna Decisions (or any two of the thirteen) in one run and says whether they agree. OpenAI Decisions API pricing, limits (128 images per request, gpt-6-luna only, no caching yet), Azure/Bedrock availability (neither offers it as of 2026-10-09) and a 13-model cost calculator: https://jev-agent.com/openai-decisions-api. This site's POST /api/v1/decisions takes OpenAI's Decisions API request with any of these model ids (OpenAI SDK with baseURL https://jev-agent.com/api/v1). - AI review analysis tool (https://jev-agent.com/ai-review-analysis): paste up to 50 reviews, name up to 8 aspects; Jev returns, per review, the probability it is positive overall and praise / complaint / no opinion per aspect with a confidence; calls under 0.5 are marked unsure; CSV export; about 1 credit a review, three runs of ten without an account. Measured 2026-10-11 on 80 MAMS restaurant reviews: 87% of labelled complaints caught, 82% of flagged complaints were complaints on a manual read, praise 57% recall; a four-way design with "neutral" scored 46% and was dropped. - Jev vs Laya (open-weight, Convai Innovations, Apache 2.0, measured by us 2026-10-09 on the same 633 items, Laya 0.4.1 English checkpoint zero-shot on an Apple M2; https://jev-agent.com/jev-vs-laya): AG News Laya 91% / Jev 88%; support tickets 74% / 100%; MASSIVE (59 options) 45% / 76%; Banking77 (77 options) 44% / 82%; toxicity 91% / 92%; positive review 67% / 98%. Wrong at 0.9+ confidence: Laya 26.8%, Jev 11.5% (Laya confidence converted to Jev's definition). Laya: 163 ms per question median on the M2. Self-hosting beats Jev on hardware cost above roughly 20 million decisions a month (400-token requests, a $0.50/hour GPU). - Gateway gotcha: Jev is a decisions model, so OpenRouter's /api/v1/chat/completions rejects it. The working endpoint is /api/alpha/decisions. It is also absent from catalogue listings because its modality is text->decisions rather than text->text, so searching a gateway for "jev" can return nothing while the model is live. Everything below in one file, for a single fetch: https://jev-agent.com/llms-full.txt ## Tools - [AI resume checker](https://jev-agent.com/ai-resume-checker): Match score, missing keywords, a recruiter's verdict. - [Essay grader](https://jev-agent.com/essay-grader): Six rubric scores and an estimated grade. - [AI grader for teachers](https://jev-agent.com/ai-grader): Your rubric, a class set, a CSV of grades. - [Rubric generator](https://jev-agent.com/rubric-generator): A rubric from your assignment, tested on an essay. - [Single point rubric](https://jev-agent.com/single-point-rubric): Template, tested example, generator. - [AI citation checker](https://jev-agent.com/ai-citation-checker): Fake references and sources that do not say it. - [AI review analysis](https://jev-agent.com/ai-review-analysis): Complaints and praise by aspect, from your reviews. - [Common App essay grader](https://jev-agent.com/common-app-essay-grader): Reflection, detail and voice, scored. - [SAT essay grader](https://jev-agent.com/sat-essay-grader): Reading, Analysis and Writing, out of 12. - [LLM cost calculator](https://jev-agent.com/llm-cost-calculator): Monthly API cost on live prices. ## Jev - [Jev AI](https://jev-agent.com/jev-ai): The model: specs, price, versions. - [Jev Router](https://jev-agent.com/jev-router): Which model it picks, measured. - [System One](https://jev-agent.com/system-one): The model class, and why it exists. - [TypeSafe AI](https://jev-agent.com/typesafe-ai): The company: funding, founders, access. - [Benchmarks](https://jev-agent.com/benchmarks): Vendor claims vs independent tests. - [Game benchmark](https://jev-agent.com/game-benchmark): 96 games, 10 it can enter, 2 it wins. - [Limitations](https://jev-agent.com/limitations): When not to use it. ## Compare - [Jev vs an LLM](https://jev-agent.com/jev-vs-llm): Same decision, both ways. - [Jev vs a classifier](https://jev-agent.com/jev-vs-classifier): When training your own still wins. - [OpenAI Decisions API](https://jev-agent.com/openai-decisions-api): Price, limits and a cost calculator. - [Decisions API vs Jev](https://jev-agent.com/decisions-api): 13 decision models, measured. - [Decision models](https://jev-agent.com/decision-models): Every model on the Decisions API, one by one. - [Jev vs Laya](https://jev-agent.com/jev-vs-laya): Hosted vs open weights, same items. - [Best LLM for classification](https://jev-agent.com/best-llm-for-text-classification): Chat LLMs and decision models, tested. - [Best LLM for sentiment](https://jev-agent.com/best-llm-for-sentiment-analysis): Reviews and emotions, 19 models. ## Examples - [Chess demo](https://jev-agent.com/chess): Watch Jev pick legal moves, live. - [Connect Four](https://jev-agent.com/connect4): Seven columns, seven live probabilities. - [Wikirace demo](https://jev-agent.com/wikirace): One Choice hop at a time. - [Agent router playground](https://jev-agent.com/jev-agent-router): Live routing, confidence policy and configuration export. - [Playground](https://jev-agent.com/playground): Run a call without a key. - [Use cases](https://jev-agent.com/use-cases): Twelve patterns, measured and runnable. - [In an agent](https://jev-agent.com/agents): Tool selection and confidence gating. - [Jev harness](https://jev-agent.com/jev-harness): Risk gate, drift alert, skills, measured. - [Jev showcase](https://jev-agent.com/showcase): Open-source projects, screenshots and demos. - [Jev demos](https://jev-agent.com/jev-demos): Community videos, set against our numbers. ## Developers - [API docs](https://jev-agent.com/docs): Calling our endpoint with your key. - [Grading API & MCP](https://jev-agent.com/essay-grading-api): The graders as an API, an MCP server and a skill. - [Jev MCP](https://jev-agent.com/jev-mcp): Judgment tools for any agent. - [Jev API](https://jev-agent.com/api-reference): How to use it, and the full reference. - [Cost & billing](https://jev-agent.com/pricing): What a decision actually costs. - [Jev API key](https://jev-agent.com/api-access): Every route to one; ours in a minute. - [Credits](https://jev-agent.com/plans): Top up when the free tier runs out. - [Get API access](https://jev-agent.com/getting-started): Waitlist and gateways. - [Jev on OpenRouter](https://jev-agent.com/jev-openrouter): Model ids, endpoint and cost. - [Rate limits & errors](https://jev-agent.com/jev-rate-limits): Every status code, tested. - [Data retention](https://jev-agent.com/jev-data-retention): Training, ZDR, DPA and hosting. - [Open source](https://jev-agent.com/open-source): Reimplementations and agents. - [Sources](https://jev-agent.com/sources): Where the facts come from. ## Worked patterns with measured output - [Support ticket triage with Jev](https://jev-agent.com/use-cases/support-ticket-triage): Ask one Choice question for the queue and one Score question for severity in the same call — both come back in roughly the time one of them would take, for a fraction of a cent. - [LLM model routing with Jev](https://jev-agent.com/use-cases/llm-model-routing): Put a Choice question in front of your model cascade so a cheap fast model handles the easy majority and only genuinely hard requests reach a frontier model. - [Agent tool selection with Jev](https://jev-agent.com/use-cases/agent-tool-selection): Let Jev choose the tool and let an LLM fill in the arguments — the selection step stops scaling badly as your tool count grows. - [Content moderation with Jev](https://jev-agent.com/use-cases/content-moderation): Use Score for severity tiers and Noul for specific policy checks, then act on the probability rather than a hard label. - [Phishing and email security classification with Jev](https://jev-agent.com/use-cases/phishing-detection): Jev's single raw verdict lost on accuracy to a cheap LLM in an independent test — but combining five cheap signal questions from the same call beat it decisively. - [RAG passage classification and reranking with Jev](https://jev-agent.com/use-cases/rag-reranking): Score each retrieved passage for actual relevance to the question and drop the ones that only matched on embedding similarity. - [Agent output guardrails with Jev](https://jev-agent.com/use-cases/agent-output-guardrails): Ask several yes/no verification questions about a draft response in one parallel call, and block or regenerate when any probability crosses your line. - [Lead scoring and qualification with Jev](https://jev-agent.com/use-cases/lead-scoring): Score fit and intent against your own rubric, and route the hot ones to a human immediately rather than in tomorrow's batch. - [Invoice classification with Jev](https://jev-agent.com/use-cases/invoice-classification): Read the invoice text once and get three typed answers: the expense category, whether a person should check it before payment, and how soon it is due. - [AI news filtering with Jev](https://jev-agent.com/use-cases/news-filtering): Give Jev a reader's interests and a headline, and get back whether to keep it, how relevant it is and what it is about, each with a probability. - [AI product categorization with Jev](https://jev-agent.com/use-cases/product-categorization): One listing in, every tag your catalogue needs out: category, main use, audience and yes-or-no attributes, each with a probability that says which tags to check. - [CSV data validation with Jev](https://jev-agent.com/use-cases/csv-data-validation): Let code check formats and types, and ask Jev what code cannot answer, row by row: is this value plausible, does this email belong to this person, what is wrong here. ## Start here - [Get API access](https://jev-agent.com/getting-started): Current key dashboards, supported gateways, and a first request. - [Official documentation](https://docs.typesafe.ai): The vendor's own docs. Where this site and the official docs disagree, the official docs are correct. ## Optional - [Sources and editorial policy](https://jev-agent.com/sources): Where every fact here comes from and how it is checked.