TypeSafe Jev
Jev is an optional decision provider from 0.22.0. It chooses from the options Plugboard supplies; chat providers still write drafts and explanations. Jev is disabled by default and has not been validated against a live school account.
Connect
Section titled “Connect”In Admin → Connectors, choose TypeSafe Jev, enter your API key and save. The saved connection is disabled. Test it with the built-in fictional example, then explicitly enable it if you want Plugboard to use it. Managed AI never adds or enables Jev. A TypeSafe account and key are your own arrangement.
Pseudonymisation is always on for decision calls, including a request that falls back from Laya to Jev. Known names and identifiers are replaced before transmission. This reduces identifying data; it is not a guarantee that every piece of free text is anonymous. Do not put unnecessary personal information into decision criteria or sample text.
TypeSafe’s standard service is hosted in the United States. Its published policy says it does not use customer requests or responses to train its models, but does not specify a fixed standard-service retention period. Enterprise zero retention is a separate arrangement. Review its privacy policy and data-processing terms before enabling.
Swap providers
Section titled “Swap providers”Laya and Jev share the same typed decision interface. Enable either, both or neither. The connector’s Decision provider order controls failover, with lower values tried first. Defaults prefer local Laya, then hosted Jev. Disable a provider to remove it from routing. If no decision provider is enabled, the engine can use your existing chat provider; those answers have no calibrated probability.
A local failure can send the masked request to an enabled hosted provider. For local-only operation, keep every hosted decision fallback disabled.
The connector requests the reviewed jev-1.13.0 version and rejects silent model
substitutions. Each version and endpoint has its own approval history. A new
model must earn automatic-action eligibility independently; safeguarding and
identity approvals still require a person.
The published price is US$0.042 per million input tokens, with output tokens free, checked on 28 September 2026. Actual billing is between your organisation and TypeSafe. Plugboard records token counts and costs, not decision text, in its AI call ledger. No live Jev benchmark has been run for this integration without an owner-supplied key.