Architecture built for enterprise reliability.

The architecture is manifest-driven end to end: service definitions, API routes, database schema, and deployment topology are all generated from declarative manifests, which is what makes four very different deployment modes — managed cloud to fully air-gapped — the same platform rather than four forks. When a capability ships, it ships everywhere, because there is one codebase and the deployment mode is configuration, not a branch.

That same discipline is why the platform stays reliable as it grows. A change to how a field is classified, a sync is scheduled, or a schema is versioned propagates through generated code, not hand-edited services — so the seventeen services can't drift out of agreement about what a connection, a mapping, or a tenant is. Every service reads the same shared manifest, and the manifest is the contract.

Protocol breadth is where integration platforms quietly draw their real boundary. Most support REST and stop, which rules out the systems enterprises actually struggle with — the payroll provider speaking SOAP, the bank speaking ISO 20022, the supply chain speaking EDIFACT, the core system speaking COBOL. Inferrex parses each of these into the same canonical model as a modern OpenAPI spec, so a mainframe copybook and a GraphQL schema become mappable to each other with no special-case machinery in between. Where no specification exists at all, live probing reads the API's actual behaviour and infers the structure from its responses.

Most fields classify without AI at all — deterministic pattern matching against The Corpus resolves them directly, so InferrexAI is reserved for the genuinely ambiguous fields. When inference does run, it runs on Inferrex's own models: a multi-layer pipeline that classifies fields by business meaning, infers relationships, and proposes cross-system mappings with confidence scores. Your corrections feed back into training, so the platform's comprehension compounds.

• Own models, own hardware — inference runs on models Inferrex trained and serves itself. No third-party model receives your schema or your data.
• Enterprise & Sovereign — inference routes within your own environment. Nothing leaves it.
• Air-gapped — private model serving with bundled weights. Zero internet dependency, full capability.
• Deterministic first — the platform prefers the corpus-backed deterministic answer wherever one exists; AI handles the residue, not the routine.

There's a question every infrastructure platform eventually gets asked: what do you support? The honest answer for Inferrex isn't a list of logos or a compatibility matrix — it's that support is a function of how you choose to deploy, not a fixed boundary on what's possible. You're not checking whether you're on the approved list. You're choosing the mode that fits.

Every API you depend on will, eventually, change its version. When it does, you're handed a bad choice. Upgrade now and hope the new version behaves — discovering the breakages in production. Or stay on the version you trust until the provider deprecates it out from under you, at which point you're forced to upgrade anyway, on their timeline, under pressure. Both options share a flaw: you find out what the new version does to your data only after you've committed to it. Inferrex removes that flaw — every version change is tested against your snapshots first, so the upgrade is a decision you make on evidence instead of a switch you flip on faith.

The full Terms, DPA and the live sub-processor list are under Legal.

Isolation enforced at the database, not in application code.

Tenant-scoped data is protected at the database layer through row-level security, with release checks designed to verify cross-tenant isolation.

Enforcing it in the database rather than in every query means a missed check is a failed release, not a silent leak.

In transit, at rest, and around credentials specifically.

External surfaces use TLS. Stored infrastructure uses platform encryption, while connection credentials are additionally protected and managed through the platform's secrets architecture.

Scoped access, yours to control where the environment demands it.

Workspace and API access is scoped. Identity and deployment requirements can vary by deployment model, including customer-controlled identity in sovereign environments.

Where inference runs, and on what.

The platform's inference architecture is designed around Inferrex-controlled deterministic and model paths, with customer structural metadata and record content treated differently.

Sovereign and air-gapped environments are designed to avoid external inference dependencies — the boundary you chose is not quietly crossed to answer a classification question.

The @inferrex/sdk package (InferrexSDK) is available. The shape:

import { Inferrex } from '@inferrex/sdk';

const client = new Inferrex({ apiKey: 'inf_live_xxxx' });

const connection = await client.connections.infer({
  url: 'https://api.stripe.com/v1',
  auth: { type: 'bearer', token: 'sk_live_xxxx' }
});

await client.connections.approve(connection.id);