What changed for Jev builders, 24–30 October 2026
This dated report records what the catalogue rechecked on 30 October 2026, after its previous checks on 24 October 2026, with no check in between. Every dated fact that a builder relies on (access, aliases, prices, routes, catalogs, lookalike sites, the hosted decision APIs compared with Jev, and pinned integrations) was either read again and re-dated, or found changed and given the date range in which the change happened. Facts that were not read again keep their earlier dates. No Jev call was made, no route was called and no account was created.
Access: nothing changed on TypeSafe's side that a new user would act on. Signups are open, new accounts get no free credit, the API is operational, jev-latest still gives jev-1.14.0, and jev-1.13.0 still has no deprecation date.
Hosted decision APIs: three facts changed for other providers (a fourth, Drex's prices, is in the table below). OpenAI's data-controls page now marks the Decisions API eligible for Private Retention and Safety Retention; Microsoft lists regions for Microsoft-Decision-1; SubQ's docs list SubQ Snap 1.1 at $0.04 per million input tokens.
Coding agents: acmelabs-15/jev-gate, a Claude Code hook, lets the tool call go ahead when Jev errors, times out after 1.5 seconds, answers malformed or has no key; by default it only logs.
When each change could have happened, 13–30 October 2026 (UTC)
- Catalogue checks
- OpenAI Decisions API: retention eligibility
- SubQ Snap 1.1 at $0.04 (docs)
- Microsoft-Decision-1 regions
- TypeSafe higher-limit request form
A bar spans the time between the last read that showed the old value and the first read, or the source's own date, that shows the new one. None of these sources gives the exact time of its change. Dots are the catalogue's checks; the hatched span is 24–30 Oct, when there was no check. Microsoft's region page carries its own date (19 Oct); the dashed line runs to the first read of it (30 Oct). The tables below give the same facts as text.
Rechecked on 30 October and unchanged
| Fact | As read on 30 Oct 2026 | Checked (UTC) |
|---|---|---|
| Signup and credit | Anyone can sign up (CEO post, 27 Sep); new accounts get no free credit. No contrary signup or credit statement dated 24–30 Oct on TypeSafe's homepage, Enterprise page, blog, Jev 1.14 post, docs quickstart or models page, the console sign-in pages or AI FrontPage (R24-S01 to R24-S03, R24-S07 to R24-S11). X could not be searched. | 05:14 |
| Service state | "All services are online". The status page's history lists no incident dated 15–30 Oct; the newest event is still the us-east-1 latency event resolved on 14 Oct (R24-S04). | 05:14 |
| Aliases and 1.13 | jev-latest and jev-preview → jev-1.14.0 ("There is no preview build available right now"); jev-1.13.0 still listed with no deprecation date; the 14 Oct post still promises "60 days warning" (R24-S05, R24-S09). | 05:14 |
| Price and context | $0.042 per million input tokens for both versions, output free; 64k per request and 32k per question by default, 262k with the long-context header (R24-S05). | 05:14 |
| Images | Images page identical to the 24 Oct copy: opt-in header and the "Dreamer-ready" note still there (R24-S06). | 05:14 |
| Six routes | TypeSafe direct, OpenRouter, Vercel, Netlify and Cloudflare list the same Jev IDs. Opper's typesafe/jev-latest still maps to jev-1.13.0. Listings only; no request was sent (R24-S22 to R24-S30). | 05:17 |
| No-account route | OpenRouter: "There's no waitlist or separate TypeSafe account" (R24-S13). | 05:14 |
| Lookalike sites | The three sites named by Eye Security still answer (HTTP 200), and the report is still online (R24-S14 to R24-S17). | 05:17 |
| Comparison-card cells | OpenAI Decisions API status and price, and its 400-questions limit; Microsoft-Decision-1 status and price; Cloudflare Clef prices (each cell on the comparison card carries its own time). | 05:17–05:19 |
| TypeSafe SDKs | @typesafe-ai/sdk 0.7.0 and typesafe-sdk 0.7.5, unchanged (R24-S35, R24-S36, R24-S39). | 05:20–05:21 |
What changed, and since when
| Fact | Before | Now | Since |
|---|---|---|---|
| OpenAI Decisions API, data retention | Not stated for the Decisions API ("Pending confirmation" on the card, 13 Oct) | OpenAI's data-controls page: "Private Retention with PSP and Safety Retention eligible: Yes, see below for limitations" and "The Decisions API supports Zero Data Retention, Private Retention with PSP, and Safety Retention"; the guide offers Zero Data Retention and HIPAA "for eligible customers" (R24-S203) | Between 13 Oct, 05:18 UTC and 30 Oct, 05:17 UTC; the row has no date |
| Microsoft-Decision-1, regions | Microsoft's region page (last updated 4 Sep) did not list Decision-1 (13 Oct) | The page, updated 19 Oct 2026, lists Microsoft-Decision-1 version 1 under Global Standard and Data Zone Standard; its Asia-Pacific tab ticks australiaeast, japaneast, japanwest, koreacentral, southeastasia and southindia. The Americas and Europe ticks were not transcribed (R24-S206) | Page dated 19 Oct 2026; first seen 30 Oct, 05:17 UTC |
| SubQ Snap, model and price | SubQ Snap at $0.05 per million input tokens (14 Oct) | SubQ's models page: "SubQ Snap 1.1" (subq-snap-1.1), public preview, $0.04 per million input tokens, output free; page modified 28 Oct. The 12 Oct launch post still says $0.05; the two have not been reconciled (R24-S42, R24-S212 to R24-S214) | Between 14 Oct and 28 Oct 2026 |
| TypeSafe, higher rate limits | A sales contact email (models page, 24 Oct) | "To request a higher limit, fill out the rate limit increase request form". Limits are still "set per organization" and shown in the developer console (R24-S05) | Between 24 Oct, 05:12 UTC and 30 Oct, 05:14 UTC |
| Drex (Nace.AI), price and signup balance | drex-max $0.06, drex-v2.0 $0.06, drex-v1.5 $0.05 per million input tokens (pricing docs, 20 Oct); "$25.00 of signup credit" (docs, 20 Oct) | The same pricing docs list drex-max at $0.03, a new drex-max-v1.1 at $0.03, drex-v2.0 $0.06 and drex-v1.5 $0.05 per million input tokens, and a signup balance of $5.00 (R24-S225). The docs page with the $25.00 wording was not re-read | Between 20 Oct and 30 Oct 2026, 05:19 UTC; the page shows no date |
| Base Labs on Baseten | Baseten's 21 Oct post described the models as free to try for a week (read 23 Oct) | The free-week wording was not found on Baseten's pricing page or its mammoth-1 page, read 30 Oct, 05:20 UTC; listed input prices per million tokens: joey-1 $0.01, elephant-1 $0.035, mammoth-1 $0.10 (R24-S223, R24-S224). The 21 Oct post was not re-read | Not dated: the two pages had not been read before 30 Oct |
None of these changes is a notice on the home page: they concern other providers, a lead, or how to ask for higher limits, not whether or how a new user can get Jev today. The three notices shown on 24 Oct are still current.
Which cloud catalogs list Jev
- Amazon Bedrock: not listed in the model cards, the Marketplace model reference or the model-choice page at 30 Oct 2026, 05:17 UTC (0 "typesafe", 0 "jev"; R24-S31).
- Microsoft Foundry: not listed on Microsoft Learn's partner-models or sold-directly pages at 30 Oct 2026, 05:17 UTC (R24-S32). The ai.azure.com model catalog page returned only a page shell without its listing, so it could not be read; it was last read on 16 Oct 2026, 05:21 UTC, when it did not list Jev.
- Cloudflare: the Workers AI models page and the Cloudflare changelog do not list Jev at 30 Oct 2026, 05:17 UTC (R24-S33); Cloudflare's AI model catalog lists
typesafe/jevandtypesafe/jev-1.14.0(first seen 21 Oct 2026; R24-S28, R24-S29).
Details: Is Jev on Amazon Bedrock or Azure AI Foundry?
Integration releases since 24 October
Registry times from npm and PyPI (R24-S35, R24-S36). Two releases were compared file by file with the version the catalogue had read; the others are listed as newer versions that were not re-read.
- ohmyjev 0.13.2 (29 Oct 2026) removes the auto-approve thresholds from its policy file and adds a lockdown setting, off by default; the 1,500 ms timeout is unchanged (package files compared with 0.12.0; R24-S38). Its failure-map row was read at an earlier version and has not been re-read at 0.13.2, so whether its cells changed is not known. Details: Claude Code and MCP.
- auto-mode 3.1.5 (29 Oct 2026): no change found in the Jev call or failure handling in the files compared with 3.0.0 (a targeted comparison, not a full re-read; R24-S37). It still requests
jev-1.13.0by ID. - Newer version exists, not re-read:
@effect/ai-typesafe4.1.0,pydantic-ai-slim2.63.0 andnace-sdk0.6.0 (framework integrations). - Patch releases:
@ai-sdk/typesafe-ai3.0.30 (29 Oct; last read at 3.0.26),@jkudish/jev-mcp0.14.7,@jkudish/jev-browser0.8.11,@openclaw/typesafe2026.10.9,litellm1.106.6 anddeepeval4.3.2; not re-read.
Coding-agent hooks: one read in source, ten classed
acmelabs-15/jev-gate (@acmelabs/jev-gate 0.2.1) is a Claude Code plugin with a PreToolUse hook on Bash, Write, Edit, MultiEdit, NotebookEdit and WebFetch. It sends each tool call to Jev with jev-latest by default. Its default mode only logs; in enforce mode it answers "ask" (the default) or "deny". Source read at commit b3ae111 (20 Oct 2026):
| Condition | What the hook does | Where |
|---|---|---|
| Jev error or timeout | fail-open the tool call goes ahead; timeout 1,500 ms | jev.ts L37, config.ts L69, hook.ts L132–149 |
| Malformed answer | fail-open a missing field counts as 0 | jev.ts L44–47, L56–62 |
| Below threshold | acts-anyway the normal permission flow continues; a call is flagged at 0.9 or above | jev.ts L69, config.ts L68 |
| No key | fail-open skipped and logged | config.ts L76–78, hook.ts L91–100 |
| Bypass | Recorded: tools outside the matcher (for example Read, Grep and MCP tools) are never scored; unparseable hook input and a missing key skip the check. All five entry points were read. | hooks.json L6, hook.ts L74–77 |
The row is not settled: what Claude Code does with the hook's answer is not public source (only its hooks reference was read; R24-S234), and in the default "ask" setting a person decides. It is shown on coding agents under "Read in source; not on the failure map yet"; the failure map still has 80 implementations. The page now counts 28 coding-agent integrations read in source: 14 act on Jev's answer, 10 do not and 4 are not settled; 27 are community projects, 1 is vendor-published and 0 come from TypeSafe. Whether anyone uses jev-gate was not checked.
Ten newer coding-agent names from 21–30 Oct were classed from their READMEs and one file listing at most (Reported, not read in source): five candidate gates, two model or effort routers, one advisory card, one triage service that is not a coding-agent hook, and one archived project, plus a personal workflow article. They are listed with their pins on coding agents.
What the scan covered
The discovery scan covered 24 Oct 2026, 05:11 UTC to 30 Oct 2026, 05:13 UTC (144 hours), joining the previous scan's end with no gap and no overlap, and was sliced by day.
- Hacker News (search): 83 items for "jev" across the six days; none for
jev-latest, "jev 1.13", "jev 1.14", "jev 1.15" or a Jev deprecation. - GitHub repository search for "jev" by creation day: 73, 69, 69, 82, 106 and 88 repositories; on 28 Oct the search returned its maximum of 100, so 6 names were not seen. Names were triaged; repository counts are not a measure of adoption.
- npm and Hugging Face searches return at most a fixed number of results, so their counts are lower bounds.
- Coverage of 24–30 Oct unknown: Reddit (HTTP 403), X (login wall; TypeSafe's own posts there could not be searched), YouTube (relative dates only), TikTok (pages rendered in the browser only), GitHub code search (needs an account), the 6 GitHub names beyond the 28 Oct limit, and the Azure AI Foundry catalog page (not readable).
Not checked or not re-read
- OpenAI's "128 images per request": the figure was not found in OpenAI's Decisions guide or API reference text on 30 Oct 2026 (R24-S46); that is not confirmed either way, so the card cell keeps its 13 Oct date.
- Microsoft-Decision-1's retention wording (the product terms were not read again), the Liquid d1-ultra line and the parts of the Drex row other than its price and signup balance on alternatives: these keep their earlier dates.
- TypeSafe's Box page, legal page and
llms.txtchanged size since 24 Oct; they were not compared line by line. The quickstart's changed lines were not identified. - ohmyjev's failure-map cells at 0.13.2; the three framework packages listed above; Claude Code's handling of hook answers.
- No request was sent to any route, so what each alias or unversioned ID returns was not observed. The catalogue has no TypeSafe account and did not observe signup.
Sources
Sources read on 30 Oct 2026 (UTC times)
- R24-S01 to R24-S13: TypeSafe console sign-in pages, the CEO post (X embed endpoint), docs quickstart, status page, models page, images page, homepage, blog index, Jev 1.14 post, Enterprise page, AI FrontPage, docs legal page and the OpenRouter Jev guide, 05:14.
- R24-S14 to R24-S17: Eye Security report and the three named lookalike sites (status only), 05:17.
- R24-S22 to R24-S30: OpenRouter, Vercel AI Gateway, Opper, Cloudflare and Netlify listings, 05:17.
- R24-S31 to R24-S33: Amazon Bedrock, Microsoft Learn Foundry and Cloudflare Workers AI pages, 05:17.
- R24-S35 to R24-S39: npm and PyPI registry documents, two package comparisons, TypeSafe SDK release feeds, 05:20–05:21.
- R24-S42, R24-S212 to R24-S214: SubQ models page, SubQ docs index and the 12 Oct launch post.
- R24-S46, R24-S200, R24-S203: OpenAI Decisions guide, API reference and data-controls page.
- R24-S205, R24-S206: Microsoft Learn pages on Microsoft-Decision-1 and its region availability.
- R24-S223, R24-S224: Baseten pricing and its mammoth-1 page.
- R24-S228 to R24-S234: jev-gate at commit b3ae111 (hooks.json, hook.ts, jev.ts, config.ts) and the Claude Code hooks reference.
Earlier dated reports: 24 Oct 2026; all reports are listed on Research.