NewRealtime voice pipeline + human-in-the-loop handoff

Put an AI agent on every call and chat

Mirakash is the operating system for AI voice & chat agents. Build one agent, connect your phone numbers and channels, and let it handle calls, chats, and messages end to end — with a human in the loop for the hard ones. One agent. Every channel. Human backup built in.

Contact centres · Support · Financial services · Education · Agencies · Multi-location

app.mirakash.com/overview
Overview /Agent overview
N
Calls today
1,284+12%
Containment
78%+4pt
Avg latency
620ms−40ms
Cost / call
$0.019on target
Live calls · now8 active
Ringing
+1 415 · billing
+1 646 · refund
+44 20 · hours
Active
+1 312 · order
+1 210 · booking
Wrap-up
+1 917 · resolved
Calls handled · 12w
Live activity
call.completed+1 415 · 2m14s
number.provisioned+1 646 555 0142
agent.publishedBilling v4
escalation.openedrefund · #4471
call.scoredCSAT 4.8
agent.startedFront desk
One platform

Replaces the patchwork most support teams run on

IVR phone tree
Vapi / Bland glue
Separate STT vendor
Separate TTS vendor
Call-recording tool
QA spreadsheet
Outbound dialer
Zendesk macros
Transcript tool
WFM staffing
Any industry

Built for every kind of customer conversation

If your team answers calls and messages, Mirakash fits. The agent, voice, channels, and workflows are configurable, so it runs the way your operation actually does.

SaaS & product support

Product teams deflecting how-to and account questions across channels.

An agent resolves order and setup questions in chat, 24/7.

Retail & e-commerce

Brands fielding order, returns, and delivery calls at volume.

An agent checks order status and books a return in seconds.

Education & admissions

Institutions handling admissions and student queries year-round.

An agent answers admissions calls and routes hot leads to staff.

Agencies & BPOs

Agencies running voice and chat on behalf of their own clients.

An agency white-labels one agent per client brand.

Financial services

Banks, lenders, and insurers with compliant, verified call flows.

An agent verifies the caller, then hands off to a human.

Logistics & field services

Dispatch, delivery, and field teams taking inbound calls.

An agent schedules a pickup and updates the delivery ETA.

Configurable core

One core, any vocabulary

Mirakash ships a small set of generic objects. Rename them, route them, and map them to how your operation actually runs — no forked codebase, no bespoke build.

Generic in, your words out
AgentVoice agent · Chat assistant · CopilotOne brain, any surface it runs on.
ConversationCall · Chat · Email · TicketAny interaction you track end to end.
ChannelPhone · WhatsApp · SMS · Web chatWhere the agent meets your customer.
WorkflowYour triggers, your stepsEvent-driven logic, not a fixed script.
HandoffTo a human, on your rulesEscalation and consult, in the data layer.
Platform

Everything an AI agent operation needs, in one system

Ten modules on one event bus and one API — from the first agent you design to the analytics on the conversation it just handled.

Build

Agent builder

Design a voice or chat agent with no code — persona, voice, knowledge, tools — and test it live in the browser.

Learn more
Talk

Realtime voice pipeline

Sub-second STT→LLM→TTS with barge-in, on ~11 pluggable provider adapters or native Gemini Live.

Learn more
Automate

Workflows & automation

An event-driven, node-graph engine that turns any signal into the right action, automatically.

Learn more
Connect

Telephony & numbers

Provision phone numbers in one click across three account models — inbound, outbound, encrypted at rest.

Learn more
Ground

Knowledge & RAG

Knowledge bases with chunking, embeddings, and retrieval so agents answer from your own docs.

Learn more
Improve

QA, scorecards & human handoff

LLM-judge scorecards, flagging rules, a review queue, and human-in-the-loop escalation that resumes the caller.

Learn more
Measure

Analytics & dashboards

Turn every conversation into insight — containment, escalation, sentiment, duration, and cost per agent.

Learn more
Reach

Omnichannel & web chat

One agent across phone, WhatsApp, SMS, email, and an embeddable, white-labeled web chat on your domain.

Learn more
Connect

API & connectors

Server actions and an API, real env-gated connectors, {{VAR}} interpolation, and agent tools that call your endpoints.

Learn more
Trust

Security & multi-tenancy

Role-based access, strict org isolation, and encrypted BYO credentials — by construction.

Learn more

See all features

Deep-dive every module, how it works, and how it connects.

Browse
How it works

From first build to a live agent, in a day

01

Build & test

Design an agent — persona, voice, knowledge, tools — and test it live in your browser.

02

Connect & launch

Attach a phone number and channels, then publish the agent in one click.

03

Automate & hand off

Workflows react to every event — call tools, branch, and hand hard cases to a human.

04

Measure & improve

Every call is scored, flagged, and rolled into analytics and cost per agent.

Automate

Turn every signal into the right action

Mirakash runs on an event bus, and Workflows is how you program it. Trigger on any event, branch on conditions, wait on timers, and fan out to actions — no code.

  • Event-driven, node-graph workflows
  • Call tools, send messages, hand off to agents
  • Actions emit events, so journeys compose
Workflows & automation
app.mirakash.com/workflows
Workflows /Workflow runs
N
RunsBuilderTriggers
TriggerIntentToolAction
Event · streamstep
call.received→ transcribe
intent.detected→ call tool
agent.stuck→ escalate to human
tool.returned→ resume speaking
silence.detected→ reprompt caller
sentiment.dropped→ soften tone
call.ended→ score + summarize
Connect

Give every agent a real phone number

Provision numbers in one click across three account models — bring your own Twilio, a managed subaccount we resell, or the platform master. Inbound and outbound, with auth tokens encrypted at rest.

  • Phone numbers provisioned per workspace
  • BYO, managed-subaccount, or platform master
  • Auth tokens AES-256-GCM encrypted at rest
Telephony & numbers
app.mirakash.com/numbers
Numbers /Phone numbers
N
NumbersProvisioningUsage
Numbers
42
Provisioning
3
Minutes · MTD
128k
GeoNumberTypeStatus
US+1 415 555 0142localActive
US+1 646 555 0188localActive
US+1 855 555 0142tollfreeActive
UK+44 20 7946 0219localProvisioning
US+1 312 555 0170localActive
US+1 628 555 0146localPorting
active · Twilio subaccount · BYO auth token AES-256-GCM encrypted at rest
Talk

One realtime pipeline for every voice agent

Speech in, speech out, in under a second. STT → LLM → TTS with barge-in, backed by ~11 pluggable provider adapters or native Gemini Live — swap any stage without touching the agent.

  • Sub-second STT → LLM → TTS with barge-in
  • ~11 provider adapters, or native Gemini Live
  • Swap any stage; the agent stays the same
Realtime voice pipeline
app.mirakash.com/calls
Conversations /Live calls
N
BoardTableLive
Agent: Front desk ▾Filter8 live · avg 1:24
Listening3
+1 415 555 0142
billing · 0:18AI
+1 646 555 0188
refund · 0:41AI
+44 20 7946 021
hours · 0:07AI
Thinking2
+1 312 555 0170
order status · 1:12AI
+1 210 555 0133
booking · 0:52AI
Speaking2
+1 917 555 0121
reschedule · 2:03AI
+1 503 555 0199
address · 1:35AI
Handoff1
+1 628 555 0146
escalated · 3:40JD
Measure

A dashboard for every agent and channel

Turn the conversation stream into insight. Define custom metrics — count, sum, unique — and compose dashboards for containment, escalation, sentiment, and cost per agent.

  • Operator-defined metrics over any event
  • Containment, escalation, sentiment, latency
  • Itemized cost + margin per agent
Analytics & dashboards
app.mirakash.com/analytics
Analytics /Support · agent analytics
N
DashboardMetricsWidgets
Containment
78%+4pt
Escalation rate
9.2%−1.3pt
Avg sentiment
+0.42+0.05
Cost / call
$0.019
Calls over time · 30dmetric: calls
Escalations · count
Cost · $/call
Recent calls
call.completed2m14s
escalationrefund
call.completed0:48
intent.detectedbooking
call.scored4.8
Improve

Score every call, escalate the hard ones

An LLM judge scores every conversation against your scorecard and flags the ones that need a look. When an agent hits its limit, it escalates to a queue that routes to a human — and resumes the caller.

  • LLM-judge scorecards, rules, and a review queue
  • Human-in-the-loop escalation to a live person
  • Consult-and-resume, not a blind transfer
QA & human handoff
app.mirakash.com/qa
QA /Review queue
N
ScorecardsEscalationsIssues
Escalations
7
Failing
2
Avg score
4.6
Resolved
92%
#c8f2
Refund not resolved
+1 646 · billing
AMescalated
#c8e9
Low CSAT · 2.1
+1 312 · order status
PKopen
#c8d4
Agent stuck
+1 628 · reschedule
JDescalated
#c8c1
Wrong store hours
+44 20 · hours
AMopen
#c8b7
Off-policy answer
+1 210 · refund
PKreview
Connect

API-first, with connectors that really fire

Drive everything from the API. Real connectors — Slack, Zendesk, Google Sheets, Notion, webhooks — execute for real, and agent tools call your own endpoints with runtime {{VAR}} interpolation.

  • Server actions + a full REST API
  • Real connectors: Slack, Zendesk, Sheets, Notion
  • Agent tools call your endpoints, env-gated
Explore the API
api.mirakash.com/docs
Integrations /API reference
N
ReferenceWebhooksKeys
POST/api/auth/token
jwt
POST/api/v2/agents
jwt
POST/api/v2/calls
key
GET/api/v2/calls/:id
jwt
POST/api/v2/messages
key
GET/openapi.json
public
POST/api/v2/callsmachine key
curl -X POST https://api.mirakash.com/api/v2/calls \
  -H "authorization: Bearer mk_live_…" \
  -H "content-type: application/json" \
  -d '{
    "agentId": "agent_billing_v4",
    "to": "+14155550142",
    "from": "+18555550142"
  }'

# → 201  call queued → streaming STT→LLM→TTS on connect
Omnichannel

One agent, on every channel your customers use

The same agent answers on phone, WhatsApp, SMS, email, and an embeddable, white-labeled web chat widget on your own domain — resolved only for verified hosts, so branding never widens access.

Omnichannel & web chat
chat.northwind.com
Northwind Support
ChatOrdersHelpContactpowered by Mirakash
<1s
Avg reply
92%
Resolved
4
Channels
Live chat
Where's my order #1042?2:14Customer
Shipped today — arrives Friday.2:14AI
Perfect, thanks!2:15Customer
Your brand · your domain · one agent across web chat, SMS & WhatsApp.
10
modules, one platform
<1s
median voice response latency
AES-256
GCM for stored credentials
agents, each isolated
Trust

Multi-tenant by construction

Isolation isn't a setting you remember to switch on — it's the architecture. The safe path is the only path.

Security & RBAC

Server-side scoping

Every query is scoped to the caller's org in the data layer — never trusted from the client.

Role-based access

Each route is guarded by a permission; the UI gates nav and actions on the same set.

Encrypted BYO credentials

Your Twilio tokens and API keys are AES-256-GCM encrypted at rest, masked on return, and never logged.

Verified-domain isolation

Custom web-chat domains and CORS both require a verified host, so branding never widens access.

The thesis
“A customer conversation is a system. It deserves an operating system — not five point tools and a call-recording spreadsheet.”

Why we built Mirakash

FAQ

Frequently asked questions

What exactly is Mirakash?+

An all-in-one platform for running AI voice and chat agents end to end: an agent builder you can test in the browser, a realtime STT → LLM → TTS voice pipeline, telephony and phone numbers, omnichannel messaging, an event-driven workflow engine, knowledge and RAG, QA scorecards, human-in-the-loop escalation, analytics with cost per agent, and a full API — all on one multi-tenant system.

How is one workspace kept from seeing another's data?+

Isolation is enforced in the data layer, not the UI. Every scoped query runs through a shared helper tied to the caller's organization, and cross-tenant access throws by construction. Any bring-your-own credential is AES-256-GCM encrypted at rest and decrypted only in-process.

Can the web chat be white-labeled?+

Yes. The embeddable web-chat widget carries your theme, logo, and custom domain, resolved only for verified hosts, so branding never widens access. The same agent also answers on phone, WhatsApp, SMS, and email.

How do connectors and the API work?+

Mirakash is API-first, and its connectors really fire — Slack, Zendesk, Google Sheets, Notion, and webhooks execute for real and are env-gated. Agent tools call your own endpoints, with runtime {{VAR}} interpolation so a URL or header can be an environment-variable decision. You can even point an AI agent at the API and have it call endpoints directly.

What happens automatically when a call comes in?+

An inbound call emits an event. The voice pipeline transcribes and responds in under a second, the agent uses its knowledge and tools to resolve the request, workflows fire on every event, and if the agent hits its limit it escalates to a human — with no manual step.

How does human-in-the-loop work?+

When an agent can't resolve something, it doesn't drop the caller. It escalates to a queue that routes to a human provider or another agent, gets an answer, and resumes the original conversation — consult-and-resume, not a blind transfer. Every escalation is tracked and assignable.

Ready to put an AI agent on every call?

Book a walkthrough and see the voice pipeline, telephony, workflows, QA, and human handoff working together — on one platform.