Jev Model Routing for AI Agents | AutoJev

Use Jev model routing to choose among approved AI models using task quality, context, tool use, cost, latency and risk requirements.

Last updated: 2026-09-18

Jev model routing selects among models your application already permits. It evaluates the task, candidate capabilities, priorities, constraints and stakes, then returns a typed recommendation that the agent can inspect before starting expensive work.

The workflow question

Which approved model can handle this task without spending more time or money than necessary?

This is different from an unrestricted model marketplace. AutoJev does not add providers or credentials. The caller supplies the allowed candidates and keeps control over permissions, budgets and fallbacks.

Example model-routing input

json

{
  "task": "Review a complex customer dispute with 100k context and tool use",
  "candidates": [
    {
      "id": "fast-model",
      "description": "Fast general model with 32k context",
      "cost": "low",
      "latency": "low"
    },
    {
      "id": "reasoning-model",
      "description": "Strong reasoning, 200k context and tool use",
      "cost": "high",
      "latency": "medium"
    }
  ],
  "priorities": ["quality", "context", "tool_use", "cost"],
  "constraints": ["Customer data must remain within approved tools"],
  "stakes": "high"
}

An illustrative result shape is:

json

{
  "recommendation": "reasoning-model",
  "confidence": 0.84,
  "escalate": false,
  "guidance": "Use the selected approved model and preserve the stated constraints."
}

Actual answers and probabilities depend on the supplied state and configured Jev provider.

When to use Jev model routing

A product has several approved models with different capabilities.
Task context, tools or risk matter more than a static default.
Routing happens frequently enough that a full generative call is wasteful.
The workflow needs a probability or escalation signal, not only a model name.

Do not use semantic routing to bypass provider allowlists, residency rules, hard budgets or deterministic capability checks.

Run the preset

Use POST /api/v1/decisions/model-route or open the model-routing example in the Playground. Agents can call the same decision through the AutoJev MCP server and the model-router Skill.

Related Jev resources

Frequently asked questions

Is this an LLM gateway?

No. AutoJev returns a routing recommendation. Your application still owns provider credentials, model invocation, retries and billing.

Can Jev choose a model that is not in the candidate list?

It should not. The candidate list is the bounded answer space supplied by the caller.

Should model routing replace hard constraints?

No. Enforce context limits, regional restrictions, permissions and budgets deterministically before or after the semantic recommendation.