Shaduf.Research preview
Jev: Use Cases, Alternatives & Products/Jev (TypeSafe AI) catalogue
Independent cataloguePage updated

Jev (TypeSafe AI): can you use it today, and where to start

Jev is TypeSafe AI's model for bounded decisions: it picks from answers you define, scores against a rubric, or gives a yes probability. This site tracks whether you can get it, what it costs, what people build with it, what the published tests show and what to use instead. (Not the Japanese encephalitis virus, JEV.)

Jev access status · checked

Available to new users: signups open, but new accounts get no free credit. API operational.

Signups open since
. Source: post on X by TypeSafe's CEO: "ANYONE CAN SIGN UP", with free credits "temporarily" disabled for new users. Documented
API operational since
, when a "System overload" event on the us-east-1 API endpoint (api.us-east-1.typesafe.ai) was resolved ("System is back to normal"). The status page shows that endpoint as "Degraded for 1 hour and 15 minutes"; its text does not give the start time (about 20:00 UTC if the stated duration ends at the resolution; an inference, not a stated time). The page's own history lists no event dated 31 Oct to 7 Nov. The event before the 30 Oct one was a 45-minute "Elevated Latency" event on the same endpoint, resolved on 14 Oct, 22:52 UTC. The main api.typesafe.ai entry's last incident was on 28 Sep (as read on 30 Oct). Status page: "All services are online" (last updated 7 Nov 2026, 15:29 UTC; read 15:29 UTC). API uptime over the last 90 days: 99.882% (the page shows three figures without labelling their windows; the third is read as the 90-day figure, as before); 99.880% on 2 Nov. Documented
Conditions for a new account
No free credit (since 27 Sep). Rate limits: TypeSafe's models page lists no numbers; it says "Rate limits are set per organization" and shows them in the developer console, which the pool cannot see without an account (until 14 Oct the page listed 100K tokens and 80 requests per second; errors and rate limits). List price $0.042 per million input tokens for both jev-1.14.0 and jev-1.13.0, output free, unchanged since 28 Sep. An opt-in header (TypeSafe-Danger-Zone-Long-Context) raises both context limits to 262k tokens at the same price per token; the defaults stay 64k per request and 32k per question. Model aliases: on TypeSafe's API jev-latest points to jev-1.14.0, unchanged since it was first seen on 24 Oct 2026, 05:12 UTC (on 23 Oct 2026, 05:45 UTC it still pointed to jev-1.13.0; TypeSafe published no dated statement of the move), and jev-preview also points to jev-1.14.0 ("There is no preview build available right now"). jev-1.13.0 is still listed with its own ID and price, and no deprecation date is published (versions and aliases). Checked 7 Nov 2026, 15:29 UTC. Documented
How it was checked
The pool did not see signup itself. On 7 Nov console.typesafe.ai answered again (HTTP 200) on / and /signup with a sign-in page ("Sign in or create an account through Google or your email") and no waitlist or invitation text. It showed a sign-in page with no waitlist text on 16, 18, 19, 20, 21, 23, 24, 30 and 31 Oct, returned an HTTP 403 block page on 2 Nov, and returned HTTP 403 until 14 Oct. No form was submitted and the pool has no account, so the status rests on TypeSafe's CEO statement (re-read on 7 Nov), its status page and its docs, not on the pool signing up.
Without a TypeSafe account
OpenRouter: "There's no waitlist or separate TypeSafe account" (OpenRouter guide, checked 7 Nov 2026, 15:29 UTC). Documented
All checks behind this status (7 Nov 2026, 15:29–15:30 UTC)
  1. console.typesafe.ai/ and /signup: HTTP 200, a sign-in page: "Welcome to TypeSafe Sign in or create an account through Google or your email", with a link to the Master Customer Agreement and the Privacy Policy; no waitlist or invitation text (on 2 Nov both returned an HTTP 403 block page). Seen as a page only: no form was submitted; the pool has no account, so signup itself was not observed.
  2. CEO post, read through X's public embed endpoint (publish.x.com, 15:29 UTC): HTTP 200, text unchanged from the 16, 18, 19, 20, 21, 23, 24, 30 and 31 Oct and 2 Nov reads. It still says "ANYONE CAN SIGN UP" and that free credits are "temporarily" disabled for new users.
  3. docs.typesafe.ai/introduction/quickstart.md: HTTP 200, 15,156 bytes, changed from the 2 Nov copy (14,948 bytes): it now shows images in the request state and lists Box as a question primitive. No waitlist, pause, credit or free-tier text.
  4. typesafe.ai: HTTP 200, no "waitlist", "credit" or "free" text found. typesafe.ai/blog (15:38 UTC): HTTP 200, eight posts. The newest, "Don't Call It a Decision Model" (dated 3 Nov 2026), is about naming and has no access, credit, price, alias or deprecation text; the newest release post is still "Jev: For the builders AND dreamers" (dated 14 Oct 2026). That post (read 15:30 UTC) is still online and still says that jev-1.13.0 "will be available for a while" and that TypeSafe intends to give "60 days warning" before jev-1.13.0 is formally deprecated.
  5. typesafe.ai/enterprise: HTTP 200, no credit, waitlist or trial text.
  6. status.typesafe.ai (15:29 UTC): HTTP 200. "All services are online", last updated 7 Nov 2026, 15:29 UTC. The page's incident history lists no event dated 31 Oct to 7 Nov; the newest entry is still the 30 Oct "System overload" on the us-east-1 endpoint, resolved 30 Oct, 21:15 UTC. The api.typesafe.ai entry shows 99.941%, 99.841% and 99.882% (2 Nov: 99.854%, 99.837% and 99.880%), without labelling their windows; the third is read as the 90-day figure, as on earlier days.
  7. docs.typesafe.ai/models.md: HTTP 200, 8,177 bytes, changed from the 2 Nov copy (8,353 bytes), but not in its aliases, model IDs, prices or rate-limit section. jev-latest and jev-preview point to jev-1.14.0, and "There is no preview build available right now" is still there. jev-1.13.0 is still listed; both versions cost $0.042 per million input tokens, output free; the long-context header applies to both; rate limits are "set per organization", and higher limits are requested through a "rate limit increase request form" (first seen 30 Oct); no deprecation date is given for jev-1.13.0. What changed between the 2 Nov and 7 Nov reads: the input row now says "Text, images" for both models (it said "Text"); a new "Question types" row lists "Choice, Score, Noul, Box" for Jev 1.14 and "Choice, Score, Noul" for Jev 1.13 ("Box questions need Jev 1.14"); the "Language support" section was removed. The page gives no date for these edits.
  8. docs.typesafe.ai/images.md: HTTP 200; the only change from the 2 Nov copy is two added "model": "jev-latest" lines in its examples.
  9. docs.typesafe.ai/changelog.md: HTTP 200, a new page (HTTP 404 on 2 Nov). Titled "Model changelog", it has one entry, "Jev 1.14", dated October 14, 2026: Box questions need Jev 1.14, and images and long context "also works on Jev 1.13". No Jev 1.15 and no deprecation are listed.
  10. AI FrontPage homepage (15:30 UTC): no credit or waitlist text found.
  11. X search: not available to the pool, so newer TypeSafe or CEO posts could not be searched.
  12. The previous check was on 2 Nov 2026, 10:47 UTC. There was no check between 2 Nov and 7 Nov 2026; for those days the status page's own history is the record (no event listed).
Access status page: timeline, incidents and details No access change since the previous check (2 Nov 2026, 10:47 UTC). Signup state, the service state at the check, price, the per-organization rate-limit wording and the alias targets (jev-latest and jev-preview give jev-1.14.0) are unchanged, and jev-1.13.0 still has no deprecation date. The status page's history lists no event between the two checks. The API uptime figure read as the 90-day window is 99.882% (99.880% on 2 Nov). The console sign-in pages could be read again (HTTP 200; a 403 block page on 2 Nov). TypeSafe's models page changed between the two checks without a date: it now lists images as an input for both models and a Box question type that needs Jev 1.14; a new docs changelog page dates its Jev 1.14 entry 14 Oct 2026.

Access for new users since launch, 15 Sep–7 Nov 2026 (latest checks 2 and 7 Nov 2026)

15 Sep1 Oct7 Nov
  • Early access from a waitlist (15–20 Sep)
  • Open to everyone (20–22 Sep)
  • Signups paused (22–27 Sep)
  • Open, no free credit for new users (from 27 Sep)

Segment lengths follow the dated TypeSafe posts on the access timeline. Existing accounts kept working during the pause. The last segment runs to 7 Nov 2026. The latest checks were on 2 and 7 Nov, with no check between 2 Nov and 7 Nov 2026; the signup and credit sources show no contrary statement dated between the checks.

Current notices

  1. Free creditDocumented

    New TypeSafe accounts get no free credit. The CEO called this temporary on 27 Sep; you pay the list price from the first request.

    Source: CEO post on X. Checked 7 Nov 2026, 15:30 UTC: no later statement restoring credit found on TypeSafe's homepage, Enterprise page, blog (one new post since 14 Oct, dated 3 Nov, about naming), the Jev 1.14 post, the docs quickstart, models or changelog page, the console sign-in page, or the AI FrontPage homepage; the CEO post is still online, its text unchanged. X could not be searched. There was no check between 2 Nov and 7 Nov 2026. Details: access status.
  2. Security warningReported

    Lookalike sites resell Jev above the list price and pass your prompts through their own servers (Eye Security, 25 Sep). Check a site before paying.

    Source: Eye Security, "Rise of the Jev-Clones". Checked 7 Nov 2026, 15:30 UTC: report online and still names all three sites; they still answer (HTTP 200). Details: official vs reseller sites.

No notice was retired or added on 7 Nov 2026 (checked 15:29 UTC); the status page lists no event between the 2 Nov and 7 Nov checks, and the models page changes found on 7 Nov are documentation edits, not current notices (they are listed under "All checks behind this status" above). No notice was retired or added on 2 Nov 2026 (checked 10:47 UTC); the status page lists no event between the 31 Oct and 2 Nov checks. On 31 Oct 2026 (checked 05:11 UTC) one notice came off this list at the end of its seven-day shelf life: the move of jev-latest to jev-1.14.0 on TypeSafe's API, first seen on 24 Oct 2026. It was not superseded: jev-latest still gives jev-1.14.0, and jev-1.13.0 is still listed with its own ID and price and no deprecation date. The fact stays on versions and aliases. No notice was added on 31 Oct; the 30 Oct status-page event was resolved before the check, so it is on the access timeline instead. No notice was retired or added on 30 Oct 2026 (checked 05:14 UTC). No notice was retired on 24 Oct 2026 (checked 05:12 UTC); one was added, the jev-latest move. No notice was retired on 23 Oct 2026. On 21 Oct 2026 two notices came off this list at the end of their seven-day shelf life (checked 05:11 UTC): the Jev 1.14 preview (jev-1.14.0 under jev-preview; at that check jev-latest still gave jev-1.13.0, and TypeSafe had published no date for the move) and the experimental image input, Box question type and 262k long context (still opt-in; TypeSafe still says images are not ready for production). Both stay on versions and aliases and limitations. No notice was retired on 18, 19 or 20 Oct 2026. On 16 Oct 2026 a notice that was not shown here, TypeSafe's Series A announcement (9 Oct), reached the end of its seven-day shelf life (checked 05:24 UTC). The post is still online; the funding announcement stays in the dated list on Research. On 14 Oct 2026 another notice that was not shown here, TypeSafe's Jack & Jill case study (7 Oct), reached the end of its shelf life; the case study stays a vendor lead on benchmarks. Earlier retired notices are listed by date on Research; their facts stay on their detail pages, such as Vercel's two Jev providers on data privacy and TypeSafe's subprocessors on the same page.

Short answer

Jev fits a repeated decision with a fixed set of answers (a queue, a label, a yes/no gate, a 2–10 level score) where your own code checks the answer and acts on it. It does not write text. TypeSafe lists it at $0.042 per million input tokens with free output (checked 7 Nov 2026, 15:29 UTC).

In published tests, all run by third parties or TypeSafe, graded studies put it ahead of small LLMs in 6 of 10, not all (benchmarks ledger); it comes close to frontier LLM judges on verdicts that can be read off the text, and behind the best LLMs on harder judgments. None of the projects this site audited is independently verified as a working production deployment.

Go straight to your task

How far this site's evidence goes

Strongest evidence held per topic, 7 November 2026

  1. TopicUnverifiedReportedDocumentedTestedBasis
  2. Access status, prices and limits by routeDocumented official statements and TypeSafe listing 7 Nov; model IDs on all six routes 14 Oct; gateway prices 8 Oct (OpenRouter 7 Oct, Netlify 28 Sep); no signup or bill observed
  3. What Jev accepts and returnsDocumented official API reference, 28 Sep; models and Images pages read 16 Oct; their text unchanged on 18, 19 and 20 Oct; on 21 and 23 Oct the models page changed size with its alias, price and limit text unchanged, and the Images opt-in note was unchanged; on 24 Oct its jev-latest row moved to jev-1.14.0, with price and limit text unchanged; on 30 Oct its alias, price and limit text was unchanged and the Images page identical; on 31 Oct the models page was identical to the 30 Oct copy and the Images page changed only in formatting markup; on 2 Nov both were identical to the 31 Oct copies; on 7 Nov the models page listed "Text, images" as the input of both versions and a Box question type on Jev 1.14 only, while the Images page still calls images not ready for production (no image was sent by the pool); Python SDK source rechecked at 0.7.4, 14 Oct
  4. Data handling by routeDocumented route and legal texts quoted 1 Oct, route labels rechecked 8 Oct, subprocessors 11 Oct; data privacy
  5. How projects and guardrails call JevDocumented source code read at pinned commits, 23 Sep–19 Oct; every failure cell recorded for the 80 implementations on when Jev fails
  6. Accuracy and calibrationReported 45 studies graded A–D on benchmarks, ledger updated 24 Oct; leaderboards listed outside the ledger
  7. Production deploymentsReported company and creator claims; none independently verified
  8. A live Jev call by this siteNone no API key; no call made

The bar fills up to the strongest level reached. Labels used on every page: Unverified lead only; Reported a named third party's own claim or measurement; Documented official documentation or source code read by the pool; Tested run by the pool (offline only so far).

What this site has not verified

  • No live Jev call, signup, account bill or credit balance. Prices are public listings.
  • No benchmark run by the pool. Test results are the operators' own and labelled Reported.
  • Search-traffic data was not available when page topics were chosen (still unavailable on 7 Nov 2026).
  • Data terms are a reading of public text, not legal advice. TypeSafe's subprocessor list was read on 4 Oct 2026 and rechecked, unchanged, on 11 Oct; its SOC 2 report sits behind "Request access" and was not seen.

What changed in each release, with dates, is on the Research page.

Search Shaduf

Search published pools, pages, reports, and evidence.