Access · Last updated 2026-09-21

Jev sandbox access and safe experiments

A Jev sandbox is a safe place to experiment: fixture tickets, a pinned model, and no bank transfers, bans, or customer emails driven by the answer. TypeSafe’s public reference does not list a separate sandbox hostname; you sandbox in your environment with a key you already have. Jev still evaluates state the same way it will in production.

This page is for builders who already heard of Jev / System One and want to try questions without creating production side effects. jev.pro is an independent unofficial field guide — not affiliated with TypeSafe AI. We do not sell, issue, or proxy API keys. Confirm hosts and model ids on docs.typesafe.ai. Keys come from the TypeSafe dashboard after access, at console.typesafe.ai — never from this site.

Steps

1. Use the official Playground to settle wording, then lock the JSON body in git. Playground clicks are not an eval harness.
2. Point the sandbox runtime at jev-1.13.0 (or the current versioned id) so later production compares apples to apples. jev-latest is fine for the first hour; pin before you fit floors.
3. Keep side effects behind a flag. Evaluation can run; approve_transfer must not. The sandbox key should not be allowed to call production tools “just this once.”
4. Prefer synthetic or redacted state. Customer PII in a shared sandbox is an access problem, not a model problem.
5. Budget tokens. Unused state still bills as input. TypeSafe’s public list price for Jev 1.13 is a vendor claim of $0.042 per million input tokens, with output free — confirm on the models page.
6. Wire the same questions module you will promote. See workspace setup and environment promotion.

Pitfalls

• Assuming TypeSafe runs a second host named “sandbox.” Public docs we used name api.typesafe.ai. Isolation is your config and your secret.
• Treating Open replicas as a Jev sandbox. OpenJev-style copies are not Jev and not endorsed. Label those results separately.
• Assuming ZDR applies because you said “sandbox.” Retention is a legal topic — read TypeSafe legal pages and data handling.
• Committing a key next to fixture JSON. Fixtures may be synthetic and committed; keys must not.
• Quoting 250k tok/s or 1200 RPM as a sandbox SLA. Those vendor claims can change and they are not a private sandbox quota.

What a sandbox is not

A sandbox is not a free tier we will invent. Confirm the models page and your contract. It is not a place to mix a TypeSafe bearer key with a Vercel AI Gateway base URL. Gateway experiments need a Gateway credential or OIDC and the model id typesafe-ai/jev.

Auth is still Authorization: Bearer or the documented gateway header. A 401 means the secret is wrong — do not retry 401. A Noul near 0.5 on fixture text is not an access failure.

When you leave the sandbox, promote the questions, floors, and version pin together. Do not promote playground edits that never landed in git.

See also

• Access hub
• Workspace setup
• Environment promotion
• Playground
• Data handling

FAQ

Is there a free sandbox tier?

We will not invent one. Confirm the models page and your contract.

Can I use OpenJev as a sandbox for Jev?

Open replicas are not Jev and not endorsed. Label those results separately.

May I commit fixture state?

Yes if it is synthetic. Never commit a key.

What this page does not claim

This page does not claim TypeSafe ships a separate sandbox hostname, a free tier, or ZDR because you used the word sandbox. It does not endorse open replicas as Jev. It does not issue keys.

Disclaimer

This is an independent unofficial site and is not affiliated with TypeSafe AI; official documentation is available at https://docs.typesafe.ai.