AI Agent Builder

Build AI agents
that handle
conversations
the way you would

Reply Flow organizes your AI like a team. An Intake Agent routes each message to the right Supervisor Agent — Sales, Support, Residents — which picks the Specialist Agent for the exact request. You define how every Specialist Agent responds, in plain English — the way you'd brief a new hire. No prompt engineering, no black box.

Build your first agent Free 7-day trial · Cancel anytime
Mira · Support Agent
3 scenarios active
  • 01

    Appointment Booking

    When someone asks about booking, scheduling, or availability…

  • 02

    Returns & Refunds

    When a customer mentions returning an item or wants a refund…

  • 03

    Escalation — Legal / VIP

    Customer mentions legal action or self-identifies as enterprise…

Active · WhatsApp + Add scenario
Three ways to start

Create an agent
your way

Whether you want a guided walkthrough, a blank canvas, or to let the AI learn from your existing conversations — there's a path that fits.

01

Guided Setup

Answer a few questions about your business — name, type, common questions, tone, escalation rules. The AI reads your answers and generates a full set of Specialist Agents automatically.

Recommended for new teams5 min setup
02

Manual Setup

Start from a blank slate. Name your Supervisor Agent, then write the trigger and instructions for each Specialist Agent yourself. Total control from day one.

For power usersMax precision
03

Import from Conversations

Upload up to 5 files of real conversation history (TXT, CSV, JSON, PDF, DOCX, XLSX — up to 10 MB each). The AI finds the patterns and generates Specialist Agents from what actually happens.

TXT / CSV / JSON / PDF / DOCX / XLSXUp to 5 files · 10 MB each
Specialist Agents

The anatomy of a
Specialist Agent

A Specialist Agent handles one kind of request, and holds the full set of instructions for it — goal, steps, guardrails, knowledge, tools, and handoff. Each is independent: edit, disable, or reorder them any time without touching the others.

Every Specialist Agent can be configured with up to 13 fields — from the trigger condition to the handoff recipient, from its tool access to a custom example reply.

In the builder today, Specialist Agents are called scenarios — the same thing, and what you'll see labelled in the screenshots on this page.

8 built-in preset templates

Product InquiryPricing QuestionComplaint / IssueShipping & OrdersHours & LocationReturns & RefundsGeneral InquiryTechnical Support
  • 01

    Scenario Name

    Appointment Booking

  • 02

    Trigger Condition

    When someone asks about booking, scheduling, or availability…

  • 03

    Goal

    Provide booking info and direct to the scheduling system

  • 04

    Step-by-Step Instructions

    1. Check availability. 2. Confirm contact details. 3. Share booking link.

  • 05

    Rules & Restrictions

    Always mention the 30-day return policy. Never discuss competitors.

  • 06

    Example Response

    Reference for tone and structure — the agent writes in this voice.

  • 07

    Pause Condition

    Customer mentions legal action or demands a refund above policy limit.

  • 08

    Communication Style Override

    Tone · Length · Emoji — per scenario, overrides agent defaults.

  • 09

    Knowledge Base Attachment

    Attach specific docs + custom instructions for this scenario only.

  • 10

    Tool Restrictions

    Choose which connectors this scenario can call — Shopify, Sheets, etc.

  • 11

    Do Not Respond

    Agent stays silent and triggers a handoff instead of replying.

  • 12

    Handoff Type

    Internal only · External to recipient · Both · None

  • 13

    Recipient Filter

    Route to Madrid team when {contact.city} = Madrid

Communication style

They sound like
you, not like a bot

Set defaults at the Supervisor Agent level — tone, length, emoji usage, and language. Then override any of these per Specialist Agent. A refund conversation can be warmer and more detailed than a quick hours-and-location reply.

Language matching is automatic — the agent replies in whatever language the customer writes in. Or you can lock it to a specific language if your team handles a single market.

Agent · Style settings Mira

Tone

ProfessionalFriendlyCasualFormal

Response Length

ConciseModerateDetailed

Emoji Usage

NoneMinimalModerate

Language

Match customerForce specific

Each Specialist Agent can override all four settings.

Fallback behavior

When no Specialist
Agent matches

Not every message fits neatly into a Specialist Agent. You choose what happens when the Supervisor Agent can't find a match — and the choice shapes how customers experience those edge cases.

AI responds normally

Default

No Specialist Agent matched, but the agent replies using default style and your general knowledge base.

AI with custom override

Configurable

No Specialist Agent matched — apply a specific tone, length, and KB instruction set just for fallback replies.

Hand off to human

Hands-off

No Specialist Agent matched — stay silent and immediately route the conversation to a team member.

Knowledge base

Answers from
your material

Drop in help docs, policy pages, product specs, FAQs — anything your team would reference. Agents answer from your material, not from the public internet.

Attach a KB to a whole Supervisor Agent or to individual Specialist Agents, each with its own custom instructions — scoping the knowledge to exactly what that specialist needs.

01

Always

Knowledge base is injected into every single message the agent processes, regardless of which Specialist Agent matched.

02

Fallback only

KB is only consulted when no Specialist Agent matches — keeps specialist replies lean and on-script.

Each KB attachment can include custom instructions — tell the agent exactly how to use the knowledge: "Prioritize the FAQ over the policy doc" or "Only cite sources when the customer asks."

Testing & debugging

See exactly
what the agent sees

No black boxes. Before you go live and while you're live, you have full visibility into how the agent is reasoning — which Specialist Agent matched, what the KB returned, what tokens were used.

Test Dialog

Type a sample customer message. See which Specialist Agent fires, the confidence level, and the full drafted reply — before you go live.

Prompt Preview

View the fully assembled system prompt for any Specialist Agent. Exactly what the model sees — no black box.

AI Debug Panel

Live in the inbox — see KB results, the matched Specialist Agent, token counts, response timing, and which model was used. Per-message transparency.

Test dialog · Mira

Customer message

"I'd like to book an appointment for next Tuesday."

Scenario match

Appointment Booking · confidence 94%

KB: 2 docs attached · 1 retrieved

Tokens: 1 240 prompt · 87 completion · 42 ms

Drafted reply

"Happy to help! We have availability Tuesday at 10 AM, 1 PM, and 3 PM. Which works best for you?"

Multi-agent routing

Two hops to the
right specialist

Every message passes through two routers. The Intake Agent reads it and picks the right Supervisor Agent — Sales, Support, Residents. That Supervisor Agent then routes to the Specialist Agent for the specific request. So a pricing question and an inventory question land in different places, even inside the same Category.

  • Intake Agent — the front desk

    Reads every incoming WhatsApp or email message and routes it to the right Supervisor Agent based on what it's about.

  • Supervisor Agents — your departments

    Each one owns an area — Sales, Support, Residents — and routes to its own Specialist Agents. Stack and reorder them; earlier ones get first pick.

  • Specialist Agents — the responders

    The matched specialist replies using its goal, instructions, guardrails, knowledge base, and connected apps — or hands off to a human.

Intake Agent · WhatsApp · incoming message

JK

"We'd like a live demo for our team of 40…"

Step 1 · pick the Supervisor Agent

Sales

Pricing, demos, upgrades

matched

Support

General

Step 2 · Sales picks a Specialist Agent

01

Demo Request

Team demos, trials, onboarding calls

matched
02

Pricing Question

03

Upgrade / Renewal

Always in your control

Edit a Specialist Agent.
It takes effect on
the next message.

No retraining. No redeployment. No waiting. Update a trigger, tighten an instruction, swap out the example response — and the change is live instantly. Your agent improves the moment you improve it.

  • Edit, disable, or reorder any Specialist Agent independently
  • Changes are live on the next inbound message — zero downtime
  • Test any change in the test dialog before exposing to customers
  • Use the prompt preview to see the exact system prompt before going live
13

configurable fields per Specialist Agent

8

preset Specialist Agent templates

3

ways to create an agent

0s

deploy time after editing

Ready to build
your first agent?

7-day free trial. Connect your channels, write your first scenarios, and have your agent answering customers within the hour.