Skip to content
VectorHub

The integration layer for the agent era

Every AI agent needs to reach real systems. Almost none of them can, because the connective tissue is rebuilt from scratch by every team. VectorHub is the layer that makes it configuration instead of engineering.

Thesis

Four claims the business rests on

If these hold, the layer gets built by someone. The question is who accumulates the catalog and the credential gravity first.

01

Capability is bottlenecked on integrations, not models

Frontier models can already reason about a refund policy or a failing deploy. What stops an agent from acting is that reaching the system holding that data is an integration project. Every capability gain in the model is capped by the connective tissue underneath it.

02

The fan-out is multiplicative, so it does not resolve on its own

N agents × M services is the shape of the problem. Teams do not grow out of it; they grow into it. Month two is when the integration backlog starts outpacing the product roadmap, and that moment arrives for everyone.

03

MCP standardised the interface but not the fan-out

The protocol was necessary and is now widely adopted. It made tools portable across clients. It did not reduce the number of servers to register or the number of credentials to distribute, which is precisely the gap a gateway fills.

04

The defensible work is the part nobody wants to do

Wrapping an endpoint is trivial. Maintaining 1,500 connectors against upstream drift, making tool schemas that autonomous callers use correctly, and running retrieval over 25,000 tools are compounding, unglamorous investments. That is what makes it a moat rather than a feature.

Product status

Live, not a deck slide

The gateway, the catalog, the converter and the discovery layer are running today in early access.

1,500+Connectors liveAcross 12 categories
25,000+Callable toolsTyped and versioned
<80msMedian gateway overheadExcluding upstream time
99.95%Target uptimeEnterprise SLA

Shipped

MCP gateway with tool-level scoping, credential vault, audit logging

Shipped

1,500+ maintained connectors across 12 categories

Shipped

OpenAPI, GraphQL and Postman to MCP conversion

Shipped

Semantic tool discovery with co-occurrence expansion

In progress

SSO, SCIM and self-hosted deployment for enterprise

In progress

SOC 2 Type II observation window

Defensibility

Why this is not a feature someone bolts on

The obvious question for any infrastructure company. Four answers, in order of how much they matter.

Catalog depth compounds

Each connector is a durable asset that gets more valuable as agents proliferate. The long tail matters more than the headline names, because the tool an agent needs is rarely the one anyone predicted.

Credential gravity

Once an organisation's OAuth grants and API credentials live in the vault and its keys are scoped, the gateway is load-bearing. Switching means re-authenticating every service by hand.

Retrieval improves with usage

Co-occurrence data from real call sequences makes discovery better in a way a competitor cannot copy from a schema dump. Working set quality is a data advantage.

Security posture as a wedge

Enterprise agent deployments stall on governance, not capability. Being the layer that answers what an agent can reach, who granted it and what it did makes procurement a reason to buy rather than a reason to wait.

Model

How it monetises

Usage-based on tool calls through the gateway, with governance features driving the move upmarket.

  • Free tier for adoption: three apps, one key, no time limit.
  • Usage-based metering on gateway tool calls; denied calls are never billed.
  • Governance features (RBAC, SSO, audit export, residency) gate the enterprise tier.
  • Self-hosted and VPC deployment as an enterprise contract, not a separate product.
  • Private connectors create per-account switching cost that grows over time.
Use of funds

Where capital goes

Catalog depth and enterprise readiness are the two things that compound. Both are hiring-constrained.

  • Connector engineering: depth and drift monitoring across the long tail.
  • Enterprise readiness: SOC 2, SSO, self-hosted deployment, security review capacity.
  • Discovery quality: retrieval research against real call-sequence data.
  • Developer relations: the adoption path runs through individual builders first.
  • Reliability: this sits in the hot path and has to behave like infrastructure.

Talk to the founders

We are happy to walk through the architecture, the connector roadmap and the numbers behind the metrics on this page. Reach out and we will send the deck and find a time.

[email protected]

See the product before the deck

The fastest way to evaluate this is to connect an app and watch an agent call it.

No credit card required · Free tier available · Self-host on request