THE ROUTING STUDIO

Decide what gets routed where.
Without writing a router.

Routing Studio is the visual editor that decides which model answers which prompt. Intent rules, confidence floors, per-mode overrides, and a live test bench, all without code.

3-panel

Editor

300+

Models

Modes

Live

Test bench

THE EDITOR

Three panels. One source of truth.

Cascade visualization on the left, rules + examples editor in the middle, and a debounced test bench on the right, every change replays against your saved snapshot in real time.

theo / routing-studio → enterprise-defaultSAVED

CASCADE

1Receive

User prompt arrives

2Resolve

CRL + MCIR context fused

3Route

Intent classifier + your rules pick the engine

4Dispatch

Provider chain (with failover)

5Audit

Decision logged with provenance

RULES

rulesexamples

vision→theo-vision≥ 95%

contains PII OR document upload

Route any prompt with PII or attached documents to the privacy-hardened vision engine.

research→theo-search≥ 90%

matches /research|cite|sources?/i

Send research-shaped prompts to the deep-search engine so you get cited sources, not hallucinations.

code→theo-core≥ 85%

matches /\.tsx?$|function |class /

Code-shaped prompts get the reasoning engine , not the conversational one.

TEST BENCH

live replay

Summarize the attached PDF (contains SSN).

expectedvision·resolvedvision

Cite three peer-reviewed sources on ICHRA renewal.

expectedresearch·resolvedresearch

Refactor this function to use async/await.

expectedcode·resolvedcode

What's the capital of Ecuador?

expectedfast·resolvedfast

All 4 tests passing • confidence floor 85%Last saved 2m ago

MEMORY VS DECISIONS

Memory Console is for facts. Routing Studio is for decisions.

The two surfaces work together: Memory Console tells you what your AI knows, Routing Studio tells you what your AI does.

MEMORY CONSOLE

The facts

What your AI remembers about each customer, account, and conversation. Source-tagged, confidence-scored, freshness-aware.

Tour the Memory Console

ROUTING STUDIO

The decisions

Which model answers, which skill fires, which guardrail trips. Visual rules, confidence floors, per-mode overrides , with a live test bench for every change.

UNDER THE HOOD

Intent classifier. Confidence floors. Failover by default.

Intent classifier

Every prompt runs through a two-tier classifier: regex fast-path in <1ms, then Haiku slow-path for ambiguous cases. Emits a confidence score per mode.

Confidence floors

Promotion to a new mode requires confidence ≥ 0.85 by default. Override per mode in the studio. The caller's preference wins if the classifier isn't certain.

Provider chain + failover

Each Theo engine has an ordered provider chain. Circuit breakers trip on > 5% error rate; the next provider takes over without the caller noticing.

STUDIO + SDK

Use the studio. Or use the API.

Every routing rule is editable straight off the API key via /api/v1/keys/[id]/routing-rules. No “preference” to create first, version control your rules alongside your code and ship them with the key.

routing-config.ts

import { Theo } from "@hitheo/sdk";

const theo = new Theo({ apiKey: process.env.THEO_KEY });

await theo.routingRules.set(process.env.KEY_ID, {
  rules: [
    { pattern: "\\b(clause|provision)\\b", target_mode: "think", confidence: 0.92 },
    { pattern: "/research|cite/i", target_mode: "research", confidence: 0.9 },
  ],
  confidence_floor_overrides: { vision: 0.7 },
});

API Reference

Take the wheel on every routing decision.

Built into every Theo account. Open an API key, write a rule, hit the test bench. No router code to maintain, no separate preference to wire up.

START FREE →OPEN THE ROUTING STUDIO

Decide what gets routed where.Without writing a router.