Access checklist
Jev API Access: Signup, Keys, Pricing & Rate Limits
Two supported ways to call Jev are a TypeSafe key and an OpenRouter key. Pick the decision model or Jev Router before you create the key. Store the key on the server.
TypeSafe direct is the native System One API. OpenRouter is the path when you already bill models through OpenRouter. Signup status below follows TypeSafe's September 27, 2026 posts. It is not a promise that the console stays open.
Two ways to get Jev API access
| Path | Use it when | Start here | JevLab note |
|---|---|---|---|
| TypeSafe direct | You want the System One API and the official SDK. | console.typesafe.ai/login | Pin jev-1.13.0 in code. Endpoint https://api.typesafe.ai/v1/systemone. |
| OpenRouter | You already have OpenRouter billing or several models. | openrouter.ai/typesafe | Choose Jev 1.13 or Jev Router first. They are different APIs. |
Other gateways also list Jev. This page covers the two paths above. Confirm model id, context, and price on the gateway you actually call.
Signup and waitlist, as of the September 27 posts
TypeSafe posted on September 27, 2026 that capacity had increased and signups were open at console.typesafe.ai/login. The same post says new signups no longer receive free credits. Diogo Almeida confirmed that day that anyone could sign up and that free credits were temporarily disabled for new users. Guides that still say signups have been paused since September 22 are behind those posts. The status can change. Check the console before you rely on it. JevLab did not complete a signup on September 29, 2026.
Posts: TypeSafe and Diogo Almeida. Last verified: September 29, 2026, against those posts. JevLab did not submit a new account that day.
After the account exists, the call sequence is:
- Create the key in that provider's console.
- Put it in a server environment variable:
TYPESAFE_API_KEYorOPENROUTER_API_KEY. - Send one smoke test with the pinned id,
jev-1.13.0ortypesafe/jev-1.13. For chat routing, the slug istypesafe/jev-router. - Set a threshold only after that call returns probabilities you can inspect. Start with the threshold guide.
API keys stay on the server
The TypeSafe SDK reads TYPESAFE_API_KEY. OpenRouter expects Authorization: Bearer with an OpenRouter key. OpenRouter's Jev tutorial says the same key bills both the Decisions API and a TypeSafe SDK pointed at OpenRouter.
- Do not put the key in browser code, a public repo, or an agent prompt.
- Do not log the Authorization header.
- If the call times out or returns an error, fail closed and hand off. A missing score is not a permit. The lab export follows that rule for its own five-label gate.
Jev pricing checked on September 29, 2026
| Surface | Model id | Input | Output | Context | Source |
|---|---|---|---|---|---|
| TypeSafe direct | jev-1.13.0 | $0.042 / 1M | $0 (not billed) | 64,000 per request; 32,000 for state plus the longest question | Models page |
| OpenRouter | typesafe/jev-1.13endpoint typesafe/jev-1.13-20260917 | $0.042 / 1M | $0 | 32,000 | Endpoints API |
Jev Router does not have this card. Its page FAQ says the listed router price is zero, and the public price feed marks the price as variable. Read that distinction on the Jev Router page before you budget a chat workload. Treat the decision-model card as the list price on the date above, and recheck it before you budget.
Jev rate limits, HTTP 429, and capacity
TypeSafe's models page lists 250,000 tokens per second and 1,200 requests per minute for Jev 1.13, and says both limits are adjusting dynamically and can change without notice. A request over either limit returns HTTP 429. The API docs also describe HTTP 529 when TypeSafe is overloaded. JevLab has not measured a queries-per-second cap.
Last verified: September 29, 2026, against docs.typesafe.ai/models and the API reference. The published pair is 250,000 tokens per second and 1,200 requests per minute. TypeSafe says higher limits are an enterprise conversation, and that the public numbers can change without notice while capacity moves. This page does not add a concurrency cap, a queue depth, or a remaining-capacity figure.
When a call is rejected or slow:
- Retry HTTP 429 and HTTP 529 with jitter. Honor Retry-After when the response sends it.
- Lower concurrency before you raise the timeout and retry the same burst.
- Fail closed into a handoff or a person. Do not invent a label because the gate had no score.
OpenRouter bills Jev 1.13 on your OpenRouter account. Its uptime fields are a provider status snapshot, not a JevLab service-level agreement. Check the dashboard you pay.
If you do not want a hosted key
A local decision model is a different system, with its own calibration. The build-your-own Jev notes compare that route with hosted Jev. They do not provision an API key.
Sources checked on September 29, 2026
Frequently Asked Questions
Is there still a Jev waitlist?
TypeSafe posted on September 27, 2026 that capacity had increased and signups were open at console.typesafe.ai/login. The same post says new signups no longer receive free credits. Diogo Almeida confirmed that day that anyone could sign up and that free credits were temporarily disabled for new users. Guides that still say signups have been paused since September 22 are behind those posts. The status can change. Check the console before you rely on it. JevLab did not complete a signup on September 29, 2026.
What is the difference between a TypeSafe key and an OpenRouter key?
A TypeSafe key calls https://api.typesafe.ai/v1/systemone and is read from TYPESAFE_API_KEY in the TypeSafe SDK. An OpenRouter key calls OpenRouter and is sent as Authorization: Bearer. OpenRouter's Jev docs say an OpenRouter key is enough for the Decisions API and for the TypeSafe SDK pointed at OpenRouter. One key does not substitute for the other.
What should I do on HTTP 429?
TypeSafe's models page lists 250,000 tokens per second and 1,200 requests per minute, and says both limits are adjusting dynamically and can change without notice. HTTP 429 means a request exceeded one of those published limits. HTTP 529 means the service is overloaded. Back off and retry with jitter. The official SDKs retry by default and honor Retry-After when it is present. Lower concurrency while limits are moving. If the call still fails, fail closed and hand the decision to a person or another path. A 429 is not a classification label. JevLab has not measured a queries-per-second cap.
Is there a free Jev tier?
TypeSafe's September 27, 2026 post says new signups do not receive the earlier free credit. Jev 1.13 is listed at $0.042 per 1 million input tokens, with output tokens not billed, on both TypeSafe's models page and OpenRouter's endpoints API as of September 29, 2026. Jev Router's OpenRouter FAQ says the router's own listed price is zero. The public price feed marks that router price as variable, and the selected model can still be billed. JevLab has not verified a permanent free tier.
Which model id should I pin for a production gate?
Pin jev-1.13.0 on TypeSafe, or typesafe/jev-1.13 on OpenRouter. Aliases such as jev-latest and ~typesafe/jev-latest can move when a new Jev release ships, which moves the score distribution your cutoff was calibrated on. Do not pin typesafe/jev-router as the classification model. Jev Router selects a chat model. It does not return the typed probabilities a gate reads.
After the key works
Put a threshold on the pinned decision model
A successful smoke test returns probabilities. The lab shows how a cutoff changes auto-adopt and handoff on one five-label fixture. It does not create an API key, and 0.80 is the slider default, not a recommended production cutoff.