Router / Triage Agent

The backend agent that reads an incoming request, classifies it, and hands it to the right specialist agent — the traffic controller of TotalApp's agent orchestration.

What Is the Router / Triage Agent?

The Router / Triage Agent is one of TotalApp's Core Agents. It reads an incoming request, classifies its intent and domain, and hands it off to the right specialist agent — a customs regulation question to the Customs Agent, a contract question to the Compliance Agent — acting as the traffic controller for the rest of TotalApp's agent orchestration.

A complex incoming request rarely announces up front which specialist should handle it. Instead of hard-coded routing rules or brittle keyword matching, the Router / Triage Agent reads what a request actually means and classifies it by intent, attaching a confidence score to every decision it makes.

In one sentence

Give the Router / Triage Agent an incoming request — it classifies it by intent, attaches a confidence score, and hands it to the destination agent with full context, or escalates it to a human triage queue if confidence is low.

How It Works — Classify, Score, Hand Off

1. Intent Classification → 2. Confidence Scoring → 3. Handoff or Escalation

Phase	What happens
1. Intent Classification	The request is classified by intent and domain, not brittle keyword rules — reading what a request actually means rather than which keywords it contains.
2. Confidence Scoring	Every routing decision carries a confidence score. Low-confidence requests are routed to a human triage queue instead of being guessed at.
3. Handoff or Escalation	High-confidence requests are handed to the destination agent with full request context — nothing needs to be re-explained downstream. Unmatched requests fall through to a general queue rather than being forced onto the wrong agent.

New agents plug in without retraining the router

Adding a new destination agent to the orchestration is a configuration change, not a retraining exercise — the router's classification approach is designed so new specialist agents can be added without disrupting how existing request types are routed.

Key Capabilities

Intent-Based Classification

Reads what a request actually means, not just which keywords it contains — surviving phrasing variation that would break a keyword-rule router.

Confidence Scoring

Every routing decision carries a confidence score — low-confidence cases go to a human rather than being silently misrouted.

Full-Context Handoff

The destination agent receives complete context — nothing is re-explained downstream, avoiding the "please repeat your issue" experience.

Configurable Per Tenant

Routing rules adapt to how each organization's teams and specialist agents are structured — one tenant's routing map doesn't have to match another's.

Output — What You Get Back

Field	Meaning
`destination_agent`	The specialist agent the request was routed to (or "unmatched" if none qualified).
`confidence`	A numeric confidence score for the routing decision.
`escalated`	Boolean — whether the request fell below the confidence threshold and was sent to human triage instead.
`context`	The full request context passed along to the destination agent.

Every decision is logged

Every routing decision — matched, escalated, or unmatched — is logged for later audit and refinement, so misrouting patterns can be caught and the routing rules improved over time.

Frequently Asked Questions

What happens when the router isn't confident about where to send a request?

It doesn't guess. Requests below the configured confidence threshold are routed to a human triage queue instead, so a low-confidence classification never silently sends a request to the wrong specialist.

Can routing rules differ between tenants?

Yes — routing is configurable per tenant, since different organizations structure their teams and specialist agents differently. There's no single global routing map.

Does adding a new specialist agent require retraining the router?

No — new destination agents are added as a configuration change. The classification approach is designed so this doesn't disrupt how existing request types are already being routed.

What happens to context when a request is handed off?

The full request context travels with the handoff, so the destination agent has everything it needs immediately — nothing is re-explained or re-requested downstream.