1. Data source
  2. Decision schema
  3. Model inference
  4. Structured result
2 Decision schema 3 typed questions
  • threatnoulprobability 0–1
  • relevancescorenot_relevantmarginally_relevanthighly_relevant
  • severitychoicelowmediumcritical

3 Model inference typesafe/jev-1.13

Waiting for a run.

4 Structured result

Run the demo to see the JSON answers, the highlighted pick and the final action.

Call Jev yourselfSame data and schema, straight to Flatkey. Set FLATKEY_API_KEY and run.
curl -sS https://router.flatkey.ai/api/alpha/decisions \
  -H "Authorization: Bearer $FLATKEY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "typesafe/jev-1.13",
  "state": "Profile: Gulf Coast semiconductor fab supply chain\nPolicy triggers: gas-turbine / power shortages, permitting moratoria, port congestion, fab tool lead times\n\nHighlights (you-search):\n- EIA: regional industrial power curtailment risk elevated this week\n- Trade press: gas-turbine OEM backlog cited as delaying two Gulf fab expansions\n- County notice: temporary moratorium process opened for large high-power facilities\n- Port authority: container dwell times +18% WoW at primary import terminal\n- Unrelated: national retail holiday spending forecast revised up\n\nTask: threat (yes/no), relevance, severity (low / medium / critical).",
  "questions": {
    "threat": {
      "type": "noul",
      "instructions": "Does this evidence indicate a material supply-chain or operational threat for the profile given its policy triggers?"
    },
    "relevance": {
      "type": "score",
      "instructions": "How relevant is this evidence to the profile and its policy triggers?",
      "criteria": [
        "not_relevant",
        "marginally_relevant",
        "highly_relevant"
      ]
    },
    "severity": {
      "type": "choice",
      "instructions": "Given the profile and surviving evidence, how severe is the current situation? Use only the fixed definitions.",
      "criteria": {
        "low": "Routine monitoring suffices; no immediate mitigation planning required",
        "medium": "Mitigation planning recommended; watch closely but not an active disruption",
        "critical": "Active disruption or clear near-term risk; immediate action needed"
      }
    }
  }
}'

No Flatkey key yet? Get one for Jev 1.13 →

Run with real Jev

Sign in and connect a key

  1. Sign in with Google.
  2. Connect your Flatkey or OpenRouter key on the next screen. Get a Flatkey key for Jev 1.13.
  3. Come straight back here and press Run again.

This site is for researching, sharing and cataloguing Jev use cases. Your key is only there to make trying Jev convenient: it is stored encrypted, used only for runs you trigger, never used for anything else, and you can delete it anytime.

About this use case

Use Jev for risk-agent triage: incoming state, Jev decision, routed outcome

Who / when

Audience

  • teams building background risk or supply-chain monitoring agents
  • harness builders who want typed judgments separate from report writing
  • ops engineers who need auditable escalate-or-skip gates under a cent

Triggers

  • a risk profile needs periodic web sweeps without a frontier model on every cycle
  • generative judges thrash across the escalate threshold run-to-run on the same day
  • you need threatProbability, relevance bands, and severity on every outcome for audit
  • deep dives should only run when triage clears a code-owned threshold

The workflow

  1. 01
    Profile + highlightsSearch hits scoped to the risk profileInput
  2. 02
    Threat?Material threat yes/no (noul)Jev
  3. 03
    Severitylow / medium / criticalJev
  4. 04
    Code gateEscalate deep dive or clean sweepYour code

Question shape

{
  "threat": {
    "type": "noul",
    "instructions": "Does this evidence indicate a material supply-chain or operational threat for the profile given its policy triggers?"
  },
  "relevance": {
    "type": "score",
    "instructions": "How relevant is this evidence to the profile and its policy triggers?",
    "criteria": [
      "not_relevant",
      "marginally_relevant",
      "highly_relevant"
    ]
  },
  "severity": {
    "type": "choice",
    "instructions": "Given the profile and surviving evidence, how severe is the current situation? Use only the fixed definitions.",
    "criteria": {
      "low": "Routine monitoring suffices; no immediate mitigation planning required",
      "medium": "Mitigation planning recommended; watch closely but not an active disruption",
      "critical": "Active disruption or clear near-term risk; immediate action needed"
    }
  }
}

Gate policy

≥ 0.90

Trust the typed answers: if threat clears your escalate threshold, run the deep dive and act on severity (low = record clean briefing; medium/critical = notify owners).

0.65–0.90

Escalate for a deep dive when threat is above threshold, but hold automated paging. Prefer a human glance at critical before any external action.

< 0.65

Treat as unclear: keep the sweep as a clean/no-action outcome, gather more highlights, and ask again before spending the deep-dive budget.

Failure modes

Pre-launch checks before you trust auto-routing:

  • Letting one frontier prompt search, judge, and write the report—so escalate-or-skip becomes a coin flip;
  • Asking Jev to write the briefing—it only returns typed threat / relevance / severity; synthesis stays with a generative model or templates;
  • Hard-coding an untested triage threshold near 0.5 without logging threatProbability on every sweep;
  • Auto-paging customers on mid-confidence critical without a human or slower review;
  • Feeding the judge empty literals instead of real profile + highlight records;
Keep exact work in code: search budgets, dedupe, URL fetch, escalate thresholds, severity → notify mapping, persistence, and MCP tool handlers are deterministic. Use Jev only for the meaning-level gates—material threat?, relevance band, severity choice—and let thresholds decide whether a deep dive runs.

FAQ

Why not let the generative model also decide escalate-or-skip?

In a measured ablation on the same profiles and news day, swapping Jev for a generative judge left report quality similar but made triage thrash across the 0.5 line—so coverage became a coin flip and judgment alone cost more than a full Jev-judged sweep. Keep generation for proposals and briefings; keep typed questions for the gates that spend money.

Where does this sit in a risk agent?

After cheap search highlights and before any deep dive. Jev answers threat (noul), relevance (score), and severity (choice); your code owns RISK_TRIAGE_THRESHOLD, query budgets, and what happens after each answer. A generative model can still propose queries and write the Markdown briefing.

What does the Flatkey call look like?

POST https://router.flatkey.ai/api/alpha/decisions with Authorization: Bearer …, Content-Type: application/json, and a body of state, model (typesafe/jev-1.13), and questions (choice / score / noul). This is not an OpenAI-compatible /v1 chat endpoint.

Can I bring my own API key?

Yes. Try the free demo above without signing in. To run it live on your own text, sign in with Google and connect a Flatkey or OpenRouter key (a quick test call runs when you save it). Manage or delete it anytime under Dashboard → BYOK. Keys are stored encrypted on the server, used only for runs you trigger, and never sent back to the browser. Need a Flatkey key? Get one for Jev 1.13 on Flatkey.

Is this the official TypeSafe Jev product page?

No. jev-llm.com is an independent research and practice site. Verify pricing, models and terms at the original TypeSafe sources.

Sources

Seen on X
Independent-site note: jev-llm.com is an independent research and practice site. It is not the TypeSafe AI official site. Verify models, pricing and terms at the original source.