Jev workflow examples
Examples on jev.pro are intent packs (triage, guardrails, ads, churn) rather than a cloned /recipes/ tree. Official cookbooks remain outbound.
Independent map — not TypeSafe’s cookbook index and not a cloned /recipes/ tree. Confirm official patterns on docs.typesafe.ai. Third-party replicas on this cluster are not Jev and not endorsed.
Start here
| Intent | Our page | Official cousin |
|---|---|---|
| Support triage | triage workflow | Intent routing pattern |
| LLM screen | guardrails | LLM guardrails cookbook |
| RAG keep/drop | passage classification | Classifying RAG passages |
| Lead composite | composite scoring | Composite scoring pattern |
How to read an example
State filter → questions → thresholds → action. If an example generates prose, it is cheating — that step is an LLM (or a template). If it computes a date or a refund amount, that step is your code. Pin jev-1.13.0 once floors exist. Log response.model. Official: adding questions on one state barely changes response time — vendor claim about fan-out, not about 10,000 rows. Bulk stays on a queue.
Anti-examples
- One Choice with 80 product questions.
- Asking Jev to write the Zendesk macro.
- Putting the whole Confluence space in
state. - Using Score expectation as a dollar refund.
If a rival recipe does any of those, we still will not.
Application names, not TypeSafe nav
Name the product surface, then pick a shape. These are application workflows, not official menu items:
SupportQueue.decideThenDraft
- Filter the ticket to subject, last customer message, product SKU.
- Jev Choice: intent (
refund/status/cancel/other) plus a Noulpolicy_ok. - Code maps to
auto/review/block(decision templates). - Only then an LLM drafts the reply.
- Optional second Jev Noul on the draft (“safe to send?”).
The LLM never chooses the refund. Jev never writes the macro. Official cookbooks describe LLM guardrails and intent routing — outbound cousins, not our nav.
LeadInbox.regexThenJev
- Deterministic filters first: empty body, known spam host, unsubscribe token.
- Remaining rows get a small Jev pack (composite Score in code — see composite scoring).
- High Score + high Noul
sales_ready→ CRM stage; middle band → human; low → drop.
Regex is not “dumb Jev.” It is the cheap gate so you do not pay tokens or trust a language judgment on a blank message. Vendor list price $0.042 / M input (confirm models page) still multiplies by whatever you stuffed into state.
RagKeep.dropPassages
Passage text in, Noul or Choice keep/drop out, then your retriever. Official cousin: classifying RAG passages. Our page: passage classification. Do not ask Jev to generate the passage.
When not to use this page’s advice
Do not paste a rival recipe and change the H1. Do not treat these names as TypeSafe product modules. Do not run the same floors on SemIf/OpenJev and report “Jev workflow accuracy.” Replicas are not Jev and not endorsed. Do not use Jev as the ESP, the bidder, or the warehouse.
What this page does not claim
- Not a full cookbook reprint.
FAQ
Can I paste a rival recipe? No. Cover the intent; write our outline.
Demos? TypeSafe hosts demos. We do not mirror them.
Is decide-then-draft official branding? No. It is our compose shorthand. Their cookbooks remain outbound.
One Choice with every product question? Anti-example. Decompose; one intent per surface.
Disclaimer
This is an independent unofficial site and is not affiliated with TypeSafe AI; official documentation is available at https://docs.typesafe.ai. Never treat jev.pro as TypeSafe official documentation. We do not sell, issue, or proxy API keys.
Open-cluster pages are independent field-guide notes. Replicas and third-party interfaces mentioned anywhere on jev.pro are not Jev and not endorsed. Hub: Open. Siblings: decision templates, usecase triage decision workflow, llm guardrails. Canonical: https://docs.typesafe.ai.