Skip to main content

ChannelHub

Your systems already have the answer.
Now your customers can get it.

ChannelHub puts AI agents on WhatsApp, web chat, avatar and API that call your existing endpoints mid-conversation — and hand the conversation to a person the moment they can’t help. Your systems of record stay where they are. Nothing migrates, nothing gets rebuilt.

WhatsApp
“Where has my order 48213 got to?”
RKCustomer · 09:41
Your API
GET /orders/48213
200 OK240 msno data copied out
Agent reply
Shipped on 12 August, arriving Thursday. Tracking is in your email.
Answered from orders API · order 48213
or it says it doesn’t know
WhatsApp
“Why was I charged twice in June?”
DLCustomer · 16:55
Your API
GET /invoices?month=06
200 OKno match
Agent reply
I can’t find a second charge in June, so I won’t guess. Passing you to Sara with the full transcript.
Handed to Sara Malik · billing
this is the point
Web chat
“Can I still return the jacket I bought?”
AMVisitor · 14:07
Your API
POST /returns/eligibility
200 OK180 mspolicy v4
Agent reply
Yes — you have until 20 August. I can start the return now if you like.
Answered from returns policy · clause 3.2
grounded in your documents
Avatar · voice
“Do you have anything Thursday afternoon?”
SBCaller · 11:22
Your API
GET /slots?day=thu
200 OK310 ms3 slots
Agent reply
Thursday at 15:00 and 16:30 are free. Shall I hold the 15:00 for you?
Answered from booking API · live availability
same agent, every channel
Web chat
“Do you ship to Oman?”
HNVisitor · 08:12
Your API
GET /shipping/zones
200 OKnot covered
Agent reply
No — we don’t ship to Oman today. I won’t pretend otherwise; I’ve logged that you asked.
Answered from shipping zones · 14 countries
a straight no beats a guess
Called as an API
POST /agent/run · “summarise account 7741 risk”
SYSYour service · automated
Your API
GET /accounts/7741/ledger
200 OK420 msread only
Agent response
Two invoices overdue by 34 and 12 days. Credit limit reached on 2 August.
Answered from ledger API · 2 records
no UI required

Integration

Governance

Connect an endpoint. Your agents call it mid-conversation, with live answers from the systems you already run.

One dashboard for every conversation

See every conversation, which systems the agents called, what they resolved and what they escalated — across WhatsApp, web chat, avatar, and API.

Studio overview: conversation volume, resolved and escalated counts, recent conversations and channel breakdown
features

Nothing on this list asks you to move your data

Point it at an endpoint you already have

Describe what a REST endpoint returns and the agent calls it when a conversation needs it. Configuration, not a bespoke integration for every question a customer might ask.

Start this week, not after the data project

Your systems of record stay exactly where they are. Nothing migrates, nothing gets rebuilt, and no team waits on a data project before you can go live.

Every answer cites where it came from

Replies are drawn from your own documents and policies, and cite them — not generated from general internet knowledge.

Explain any answer, months later

Every routing decision, retrieval and tool call is recorded, so you can account for how any answer was produced.

Never evaluate a model again

We select the model best suited to each use case and route the work to it. You don't evaluate models, hold provider accounts, or revisit the decision every time a better one ships.

One source of truth, however many agents

Purpose-built agents for distinct jobs, sharing one source of truth instead of drifting apart in separate tools.

Your team picks up mid-conversation

A shared queue for escalations — your team picks up with the full conversation context already attached, so the customer never repeats themselves.

See exactly where a person was still needed

How much the agents resolve on their own, which systems they called, and exactly when a person had to step in.
why we built it this way

We got tired of AI that always had something to say

Cites its source — or says it doesn’t know.

Every answer carries the document it came from. When there isn’t one, the agent says so and hands the conversation to a person with the transcript attached.

We kept seeing the same failure mode in AI tools: a system that always has something to say, even when it shouldn’t. So we built ChannelHub around the opposite instinct — pick the right model for the job, answer from real data, say so when it doesn’t know, and get a person involved cleanly instead of guessing.

The ChannelHub team
Built by Codelabs

Start with one endpoint

Pick the question your team answers most often, point an agent at the system that already knows, and watch what it does with the ones it can’t. That’s the whole evaluation.

Three days free. A card is required to start and is not charged during the trial. When it ends you move to Starter at $49/mo unless you cancel first — we email you the day before either way. The avatar channel is not included.

Not ready to connect an endpoint yet? Tell us what you’d like to connect using the form at the foot of this page, and a person will reply — usually within one working day.