Hermes + Jev

A redacted guide to how an agent workflow can use TypeSafe Jev for typed semantic judgments while deterministic code and human approval retain control of outcomes.

Choose a view

• ArchitectureComponents, trust boundaries, local credential isolation, and the external AI service boundary.
• Data flowHow sanitized context becomes a typed judgment and then a policy-gated outcome.
• Request sequenceThe request/response exchange, including local-only credential handling and human review.
• Decision lifecycleWhen Jev is applicable, when deterministic handling is enough, and when review is required.