Latest intelligence ↗
PHASE 05 / INTELLIGENT ROUTER

Right request.
Right model.

Stop paying frontier prices for every task—and stop routing blindly.

Build an ordered policy, simulate every decision locally, enforce request budgets, recover through explicit fallbacks, and export the complete production configuration.

01CLASSIFY

Describe the task, request size, and constraints that should activate a route.

02ROUTE

Choose the first appropriate model using ordered, inspectable rules.

03GUARD

Block estimates above the policy cost ceiling before live execution.

04RECOVER

Attempt explicit fallbacks only when the selected route fails.

05EXPORT

Take the complete policy into an application as JSON or TypeScript.

01 / POLICY

Define the operating limits.

Rules run from top to bottom. The first complete match selects the primary model; the fallback chain is used only when execution fails.

ORDERED RULES

3 decision gates

01
02
03
FALLBACK CHAIN

Ordered recovery routes after a provider or model failure.

02 / REQUEST SIMULATOR

See the decision before spending a token.

Simulation is deterministic and local. It shows every rule checked, the reason for selection, the fallback chain, and an estimate based on the configured token ceiling.

SELECTED PRIMARY

Nemotron 3.5 Lightning

A compact free route is sufficient for bounded classification.

WITHIN BUDGET
$0.00000ESTIMATED COST
528ESTIMATED TOKENS
100%VS GROK 4.6 ESTIMATE
1FALLBACK ROUTES
DECISION TRACE
✓Fast classification

Matched: classify

—Software work

Skipped · task

—Complex reasoning

Skipped · task

✓Budget guard

$0.00000 estimated / $0.025 maximum

→Final route

Nemotron 3.5 Lightning

03 / LIVE DISPATCH

Run the policy, including fallbacks.

The router calls only the selected primary. A fallback is attempted only when the previous route returns an execution failure.

OPENROUTER KEY REQUIREDHeld in this tab only; never saved by AXON//RADAR.
04 / PRODUCTION HANDOFF

Take the policy with you.

Export the entire ordered policy as portable JSON or a typed TypeScript configuration.

{
  "version": 1,
  "kind": "axon-radar-routing-policy",
  "id": "balanced",
  "name": "Balanced production router",
  "description": "Use compact models for routine work and escalate complex requests.",
  "maxCost": 0.025,
  "maxLatency": 12000,
  "defaultModel": "google/gemini-3.7-flash",
  "fallbacks": [
    "z-ai/glm-5.3",
    "nvidia/nemotron-3.5-lightning:free"
  ],
  "createdAt": "2026-08-20T00:00:00.000Z",
  "rules": [
    {
      "id": "classify",
      "name": "Fast classification",
      "enabled": true,
      "task": "classification",
      "keywords": [
        "classify",
        "label",
        "category",
        "sentiment"
      ],
      "minChars": 0,
      "maxChars": 4000,
      "modelId": "nvidia/nemotron-3.5-lightning:free",
      "reason": "A compact free route is sufficient for bounded classification."
    },
    {
      "id": "code",
      "name": "Software work",
      "enabled": true,
      "task": "coding",
      "keywords": [
        "code",
        "function",
        "debug",
        "sql",
        "typescript",
        "python"
      ],
      "minChars": 0,
      "maxChars": 12000,
      "modelId": "deepseek/deepseek-v4-pro-0813",
      "reason": "Route implementation and debugging work to the open coding specialist."
    },
    {
      "id": "complex",
      "name": "Complex reasoning",
      "enabled": true,
      "task": "reasoning",
      "keywords": [
        "analyze",
        "strategy",
        "trade-off",
        "architecture",
        "prove"
      ],
      "minChars": 800,
      "maxChars": 12000,
      "modelId": "google/gemini-3.7-flash",
      "reason": "Escalate longer analytical work to the frontier reasoning route."
    }
  ]
}
LOCAL WORKSPACE

Saved policies and recent dispatches

POLICIES

No saved policies yet.

LIVE RUNS

No live dispatches yet.