What latency can we actually expect?

Median STT→LLM→TTS p50 stays under 800ms on standard cloud infra; on-prem with a small LLM hits ~400ms. Numbers vary with model size and language; we'll benchmark on your audio in the trial.

How does it work for non-English calls?

32 languages today. STT and TTS share a language graph so code-switching mid-call (e.g. EN-TR) keeps tools working. Quality grades vary by language — we'll send you our latest WER table.

Yes. The full stack runs in your VPC or on-prem on commodity GPU hosts. Same APIs and dashboards as the managed cloud. Air-gapped deployments are supported on the enterprise plan.

What if our tool isn't on the integrations list?

If it has an HTTP API or webhook, we can wire it up the same day. Native integrations are first-class; everything else is a JSON-schema tool the assistant can call mid-conversation.

How do you handle PII and recordings?

Recordings and transcripts are encrypted at rest. Optional auto-redact removes PII before storage. You control retention windows. BYOK with your KMS is available on enterprise.

How is pricing structured?

Per-minute base + per-tool-call. No per-seat fees. We don't publish numbers because volume and self-host change the math — talk to sales for a quote tailored to your traffic.

Do you train models on our calls?

No. Your audio and transcripts are not used to train shared models. Optional opt-in is available if you want a tuned model trained only on your data, kept private to your tenant.

Real-time analytics, native warm-transfer to human queues, an open-source assistant SDK, and richer eval tooling for regulated workloads. Public roadmap on request.

One platform, every channel

The enterprise AI assistant platform.

It works with your documents, runs your processes, and stays the same assistant on every channel. Whatever your industry, you shape the assistant with your own knowledge.

Answers from your documents, with sources
Web, email, phone, and integrations
Cloud or on your own servers

What you get on day one.

12 platform capabilities and 12 ready-made connections — already wired, just point them at your sources.

Answers from your documents

Source citations

Step-by-step flows

Tool calling

Scheduled tasks

Model selection

Multilingual

Voice conversations

Phone line

Handoff to a human

Personal-data redaction

Audit trail

Answers from your documents

Source citations

Step-by-step flows

Tool calling

Scheduled tasks

Model selection

Multilingual

Voice conversations

Phone line

Handoff to a human

Personal-data redaction

Audit trail

HHubSpot

SSalesforce

ZZapier

SSlack

GGoogle Calendar

TTwilio

SStripe

NNotion

IIntercom

PPipedrive

MMake

ZZendesk

HHubSpot

SSalesforce

ZZapier

SSlack

GGoogle Calendar

TTwilio

SStripe

NNotion

IIntercom

PPipedrive

MMake

ZZendesk

Live demo · Inbound

Watch an assistant handle a real call.

From the first ring to the CRM update — every step happens in seconds. The full transcript is captured and a human can take over at any moment.

1Caller dials in
2Speech transcribed in real time
3Intent classified
4Appointment booked
5CRM record updated
6Confirmation SMS sent

Inbound · Live

Live transcript

Live demo · Web chat (chatbot)

Watch a question on your site become a cited answer and an open ticket.

The visitor types → the assistant answers from your help document and shows the source → pulls the order status from your system → hands what it can't solve to a person, history included.

1The visitor types on your site
2The assistant answers from your document, with the source
3Pulls the order status from your system
4Hands off to a person, history included

Web chat

How it works

A working assistant in three steps.

Connect your knowledge and your apps, set what the assistant does, then publish. Every step is reversible — you change the live assistant without stopping it.

Connect

Bring your documents, your tables, and the apps you already use — add your phone line too, if you want one. Calleague speaks to the sources your team uses today.

Configure

Choose what the assistant does, which tools it may call, its language, and when it hands off to a human. Test it with real questions before you publish.

Go live

Open one channel first, then roll out to the rest. Watch every conversation live — full text, the sources it used, and the steps it ran.

Live demo · Outbound

Run a campaign across thousands of numbers — and watch it land.

Upload your contact list, set your script, and Calleague dials in parallel — handing off to a human the moment it matters.

N parallel call lines — no per-line licensing, no per-seat fees.
Voicemail detection drops to a leave-message branch automatically.
Hot handoff to a human agent on intent match, sentiment shift, or callback.

Outbound · Campaign

Dialer matrix0%

AnsweredVoicemailNo answerDialing

Answered

Booked

Voicemail

No answer

Four building blocks

The four building blocks an enterprise assistant needs.

Each block earns its keep alone, and all four work together. Bring your own models, your own sources, and your own servers.

Flow orchestration

The assistant does more than reply: it runs steps in order, branches on conditions, calls tools, and writes the result back where it belongs. Scheduled tasks run those same flows without being asked.

Answers from your documents

Upload your documents, tables, and web pages; the assistant grounds every answer in them and shows which passage it used. Update the source and the answer follows.

Model governance

You decide which job runs on which model. Per-team and per-user permissions, usage limits, and cost live on one screen — cloud providers and models on your own servers sit in the same list.

Enterprise architecture

Teams, roles, and fine-grained permissions; secrets encrypted at rest, an audit trail, and single sign-on. Run it in our cloud or entirely on your own servers — your data stays yours.

One conversation that picks up where it left off — across channels.

The customer types on your site, replies by email, then picks up the phone — the assistant sees the same history in all three, and hands the whole history over when a person takes it. The phone line is just one of these channels.

Web chat
Embed on your site
Email
Phone line
Integrations
Desktop & Office

The customer types on your site
Replies by email — the assistant sees the same history
Picks up the phone — the conversation continues
Handoff to a person — full history included

one conversation history

Architecture

Six layers, all visible.

Calleague keeps the plumbing visible. Every layer can be swapped, scaled, or replaced — and every step is observable end-to-end.

Sub-second latency

Median STT→LLM→TTS round-trip stays under 800 ms on standard infra (p50).

32 languages

STT and TTS share the same language graph, so code-switching mid-call doesn't break tools.

Self-host or SaaS

Run the full stack in your VPC, on-prem, or use our managed cloud. Same APIs, same observability.

Live demo · Integrations

Watch a real call become a CRM record, a calendar slot, and an SMS.

Every step is observable in the dashboard. The packet labels are real payload field names.

2,400,000+

Conversation minutes processed

Across pilots, betas and self-hosted deployments to date; mostly voice calls.

Languages supported

STT and TTS coverage across the same language graph.

<800 ms

Round-trip latency (p50)

STT → LLM → TTS, under 800 ms on standard infra.

99.95%

Last 90 days availability

Not an SLA commitment; talk to sales for a contractual SLA.

Numbers reflect aggregated platform telemetry. Pilot and self-hosted figures vary; ask us for the deck.

Security

Built for regulated, voice-of-the-customer workloads.

Voice data is sensitive. Calleague encrypts it everywhere, lets you keep it where you need to, and proves it's handled the way you say it should be.

Encrypted in transit & at rest

TLS 1.3 between every hop. AES-256 for recordings and transcripts at rest. Your keys, your KMS, optional.

Data residency

Pin recordings, transcripts, and metadata to a region (EU, US, TR). On-prem keeps every byte inside your network.

Compliance-ready

Designed to SOC 2 Type II controls and to meet GDPR and KVKK requirements. HIPAA deployments are available with a BAA on the enterprise plan.

SOC 2
GDPR
KVKK
HIPAA — with BAA

FAQ

What teams ask before they buy.

Straight answers. If you don't see yours here, ask sales.

Median STT→LLM→TTS p50 stays under 800ms on standard cloud infra; on-prem with a small LLM hits ~400ms. Numbers vary with model size and language; we'll benchmark on your audio in the trial.

Build your first assistant today.

Upload a few documents, pick a channel, and test it with real questions. A dedicated team is not required to set it up.