GUIDES

Agent intent router boundary tester and LLM cost calculator

Stress-test your intent labels against negation traps, conflicting multi-action prompts, and embedded instructions before wiring an AI agent, and calculate how much frontier LLM spend a System-1 router saves.

JEV / AGENT INTENT ROUTER TESTERNEGATION · MULTI-INTENT · INJECTION GUARD

Route: invoice · Simulated confidence: 0.94 · needsReview: false — Negation respected.

100% Frontier LLM: $540/mo → Jev System-1 Router + Selective LLM: $471.80/mo.

SIMULATED ROUTER OUTPUT & TYPESCRIPT MIDDLEWARE
{
  "id": "dec_sim_44109",
  "model": "typesafe/jev-1.13",
  "choice": "invoice",
  "confidence": 0.94,
  "probabilities": {
    "invoice": 0.94,
    "cancellation": 0.06
  },
  "needsReview": false,
  "replayed": false
}
Run live in Agent Runner ↗

const response = await fetch('https://jevapi.pro/api/v1/decisions', {
  method: 'POST',
  headers: {
    'Authorization': 'Bearer ' + process.env.JEV_API_KEY,
    'Content-Type': 'application/json',
    'Idempotency-Key': 'router-turn-0001',
  },
  body: JSON.stringify(payload),
});
const decision = await response.json();
if (decision.needsReview) return escalateToHuman(decision);

Ready to run this decision schema against live Jev 1.13?

Start with a one-off $29 Top-up (1,800 credits, never expires) or choose Builder ($39 for 30 days · 3,500 credits) and mint a server-side API key in 30 seconds.

Label a held-out set, compare model decisions with it, and review errors before changing a workflow.
Label a held-out set, compare model decisions with it, and review errors before changing a workflow.

Why naive chat routers fail on negation and mixed intents

Keyword rules and uncalibrated chat prompts routinely misroute messages like "Do not cancel my account; I only need an October invoice" because the word cancel dominates attention. Worse, when a user asks to cancel a plan and dispute a charge in the same turn, forcing a single action without a clarify route triggers expensive mistakes.

Click the four boundary presets above to inspect how an explicit clarify outcome paired with a strict confidence cutoff keeps ambiguous and adversarial inputs out of automated tool execution.

Model and task routing guide · Try live presets in Agent Runner

Put a System-1 decision gate ahead of expensive reasoning models

Sending every inbound support or agent turn straight to a heavy chat model burns thousands of prompt tokens on routine invoice lookups, status checks, and FAQ deflections. Placing Jev 1.13 at the ingress classifies bounded turns in one lightweight call and reserves frontier reasoning models for complex synthesis.

Ingress routing architecture comparison
ArchitectureBounded answer domainUncertain turn handlingCost profile at 30k turns/mo
100% Frontier Chat LLMRequires JSON schema + retry parserSilent guess unless custom rubric addedHigh token bill on every turn
Regex / Keyword RouterExact strings onlyBreaks on negation ("do not cancel")Cheap but high misroute toil
Jev System-1 Router + FallbackStrict 2–12 allowed labelsExplicit needsReview flag + clarify routeFlat 1 credit/turn ($299 Growth = 50k)

Cohere Classify alternatives · Compare API plans

Keep authorization and destructive actions outside the router

Even at 0.99 confidence, an intent label is a routing suggestion, never proof of identity or permission. Always verify session ownership and require explicit user confirmation before executing cancellations, refunds, or account changes. Ready to wire your router? Grab a $29 Top-up or $39 Builder API key in seconds.

Build custom JSON schemas · Get your Builder API key

Agent intent routing FAQ

Should my agent execute a cancellation automatically when choice is cancellation?

No. Treat the router choice as queue or workflow selection; your application code must still verify authentication, permissions, and explicit user confirmation.

How do I connect Jev directly to an MCP-compatible coding or support agent?

Point your MCP client at https://jevapi.pro/api/mcp with a workspace Bearer key to expose six stateless decision tools without custom glue code.

Try a decision ↗

Continue your workflow

Guides

An LLM router that returns a decision

Routing is a bounded choice: which of the models or paths you actually have should handle this task. Write the criteria down, let a decision model choose among your candidates, and keep the call itself in your code.

Read the guide ↗
DAILY BONUS

2 free credits, every day

Claim 2 free credits each UTC day — 14 across seven claims. No purchase needed. Credits do not expire.

    Resets at 00:00 UTC. No streak required.

    Try a decision →

    Take the next decision into your workspace.

    1 anonymous attempt per day · 20 signup credits · No card for the trial

    Sign in ↗

    Tell us what you need

    Describe the workflow you want to classify, the volume you expect, and the outcome you need. We reply by email.

    10–2,000 characters. Never include passwords, API keys, or payment details.