Anatomy of an Agent Stack
An Agent Stack is not a single prompt with a giant system message. It’s a layered system, and the layers do different jobs. Once you’ve seen them, the rest of the configuration UI starts to make sense.
The three runtime layers
Customer message
│
▼
┌───────┐
│ ROUTER │ Picks the right specialist for this message
└───┬────┘
│
▼
┌──────────────┐
│ ORCHESTRATOR │ Watches the conversation, intervenes when needed
└──────┬───────┘
│
▼
┌────────────────┐
│ SPECIALIST │ Replies to the customer
└────────────────┘
Router
The router is the front door. When a customer message arrives, the router decides which specialist should handle the topic. It does this with LLM-based topic matching against each specialist’s scope (responsibility text) and any explicit routing topics defined for that specialist.
You can customize the router’s prompt from the Orchestrator tab — there’s a free-form text box where you append rules to the built-in defaults, like “for refund questions, prefer the Billing specialist over the Tech specialist.” Anything you write here is added to the default; the built-in actions and core rules always stay in effect.
Orchestrator
The orchestrator is the manager. It runs a state machine over every conversation and handles:
- Stall detection — when a specialist is going in circles, the orchestrator steps in.
- Re-routing — if the wrong specialist was picked, it can hand the conversation to a different one without the customer noticing.
- Risk escalation — if the customer signals frustration, legal threat, churn risk, or asks to escalate, it triggers handover.
- State management — tracks where every conversation is in its lifecycle.
You control the orchestrator’s check interval, stall threshold, risk auto-escalation, and handoff mode. See The orchestrator and the orchestrator-states reference.
Specialist agents
Specialists are the experts. Each one is a separate config with its own:
- Scope (responsibility) — the kind of questions it handles. Used by the router.
- Instructions — specialist-specific behavior, required steps, escalation triggers.
- Knowledge access — which Knowledge Base categories and Handbook procedures it can use.
- Capabilities — the actions it can take in your other systems.
- Playbooks — multi-step procedures the specialist follows for specific tasks.
A specialist’s reply temperature is forced to 0 at runtime — specialists are precise, not creative. Stack-wide tone and voice live on the Personality tab; specialist-specific behavior lives in Instructions. See What is a specialist agent?.
Personality is stack-wide
Tone, voice, formality, warmth, humor, and emoji policy all live on the stack’s Personality tab — they’re shared by every specialist in the stack. You pick a preset (Friendly Helpdesk, Expert Advisor, Concierge, Efficient, Brand Playful) or “None” and tune the dimensions from there. See the personality presets reference.
This split exists for a reason: customers should not feel like they’ve been transferred to a different person every time the router hands off. Tone stays consistent across the stack; expertise changes between specialists.
Background — infrastructure agents
In addition to the three runtime layers, every stack runs four infrastructure agents in the background. They handle cross-cutting concerns:
- Conversation summary — Generates a running summary of the conversation, used at handoff and shown in Sidekick
- Sentiment & risk — Tracks customer sentiment and flags conversations as risky for auto-escalation
- Handoff pack — Prepares a briefing — summary, sentiment, suggestions — for the human picking up
- Missing info — Identifies what information the AI still needs to resolve the issue, and prompts the customer for it
Each can be toggled on or off and has its own configuration. See Infrastructure agents reference.
Tabs in the configuration UI
When you open a stack from Settings → Agent Stacks, you see these tabs:
- General — Name, description, avatar, enabled toggle, auto-close inactivity, handover-when-unsure
- Personality — Persona presets and tone dimensions for the whole stack
- Orchestrator — The stack diagram. Click the orchestrator to customize routing rules; click any specialist to drill in
- Agents — Flat list of specialists with the same drilldown (legacy view; the Orchestrator canvas folds this in)
- Handover — Triggers, prerequisites, delivery — when, what, and to whom
- Voice — Voice-specific config (only visible when the AI voice agent flag is on)
- Testing — Test sandbox for the stack and individual specialists, plus the tuning flow
- Change History — Audit log of changes to the stack
- Danger Zone — Delete the stack
Inside any specialist drilldown you’ll find four sub-tabs: Setup (name, scope, instructions), Capabilities (which tools it can call), Knowledge (which KB categories and Handbook procedures it can read), Playbooks (multi-step procedures).