Embedded agent infrastructure

Ship AI agents inside your product

Your users build them. Your brand, your UI. We run the infrastructure.

The panel above is GTWY. Your users only ever see your product.

400+ business customers building agents today4 lakh runs per monthLive in production across ViaSocket and MSG91

Where this starts

You already know you need this.

Your roadmap says “add AI.”

Your users don’t want another chatbot bolted on the side. They want agents inside your product, configured for the way they work.

Building one agent is easy. Building an agent platform is not.

Isolation per customer, scoped tokens, per-customer usage, theming, a builder UI your users can actually operate — nine months of work nobody on your team wants to own.

Meanwhile your competitors are shipping.

The window on this closes at the speed of whoever moves first in your category.

That’s the part we’ve already built.

How it works

Four steps, and your users are building.

1Add the embed to your product
2Your customers build agents in your UI
3Agents run on our infrastructure
4You get per-customer usage to bill against

See the full integration →

Integration walkthroughVideo asset pending — GTWY-816

Full integration, start to finish — under 30 minutes.

What you get

Everything the platform version needs.

Every customer isolated

Their agents, their data, invisible to everyone else.

Data isolation →

Your brand, not ours

Your logo, colours, domain. Your users never see GTWY.

White-label →

2,000+ integrations

Agents that act, not just answer — in Slack, HubSpot, Sheets, Zendesk and two thousand more, available to your users on day one.

Browse integrations →

Templates so builders aren’t staring at a blank canvas

Working agents your users can clone on day one.

Templates →

Reference architecture

Running at scale, today.

Two platforms in the Walkover group run their agent features on this infrastructure. Their architecture is the reference for yours.

ViaSocket

Workflow automation platform

What they needed: An AI step their own users could configure — not a fixed feature, a builder. Every one of their customers needed their own agents, keys and data.
Why they didn’t build it: The agent runtime was three months. The multi-tenant layer around it — isolation, scoped tokens, per-customer metering, a themed builder — was the rest of the year.
How it’s integrated: The embed renders inside their workflow editor as an AI step. One token minted per customer at session start; usage read back per customer for billing.
Volume handled: 400+ business customers building agents, 4 lakh runs per month.
What their customers build: Lead qualification, ticket triage, document extraction, CRM enrichment — each one wired to the tools that customer already uses.

Read the full write-up →

MSG91

Communication APIs

Agents that answer and act on customer conversations, embedded in the MSG91 console under their own brand, isolated per account.

Read the write-up →

Pricing

Pay for what your customers run.

Usage

$0.06 per run. LLM costs passed through at cost.

Enterprise

Volume pricing, dedicated infrastructure, custom terms.

See pricing in full →

Questions

The ones you’re already asking.

How long does integration actually take?

Configuration in our panel takes 15–30 minutes. The snippet itself is a few lines. Most teams are running against staging the same day and spend the rest of the week on their own UI decisions.

Can our customers see each other’s data?

No. Each of your customers gets an isolated environment, and a token is valid for exactly one of them. A token from customer A cannot read customer B’s agents, documents or run history.

Does it look like our product?

Your logo, colours, typography and domain. There is no GTWY logo, no “powered by,” and no GTWY domain visible to your users.

What models are supported?

The major hosted providers, with fallback routing between them. You choose which models your customers are allowed to select, and a provider outage degrades a run rather than taking down your feature.

How do we bill our own customers for this?

The usage API returns runs attributed to each of your customers. Charge per run, bundle it into a plan, or fold it into a premium tier — the spread between what you pay and what you charge is yours.

What happens if we outgrow it, or want to leave?

Agent configurations and knowledge sources are exportable per customer. Enterprise agreements cover dedicated infrastructure and exit terms in writing.