---
title: "Integration & Gateways"
description: "AS2 gateway for EDI partners, cloud storage, file service, natural-language AI assistant."
canonical_url: "https://preview.openepcis.io/docs/platform-overview/modules/integration"
last_updated: "2026-07-30T18:34:19.509Z"
---

This is the perimeter — the place where OpenEPCIS meets the rest of the enterprise. Trading partners send EDI over AS2. Older systems drop files into an S3-compatible bucket or Azure Blob. Operators ask the AI assistant natural-language questions about what's in the event store. All of this lives in the Business edition.

The AS2 gateway is the busiest of the bunch in practice. Partners send EPCIS-XML envelopes, the gateway validates them, hands them off to `/capture`, and the MDN response that goes back includes the EPCIS capture-job ID — so AS2-level acknowledgement and EPCIS-level processing stay reconciled without anyone having to chase a correlation by email. Partner certificates and trading-relationship metadata are database-backed, so onboarding a new partner is a REST call, and there's a self-service enrollment flow for partners who'd rather not wait for a manual setup.

Object storage is pluggable. Any S3-compatible store and Azure Blob are first-class options, so the same captured EPCIS document can be archived to whatever the customer's environment already uses — AWS S3, on-prem object storage, Azure — without standing up a separate bucket or rewriting the capture pipeline.

The AI assistant is intent-routed rather than free-form. A prompt is classified into one of a small set of intents — EPCIS query, Digital Link resolution, vocabulary lookup, general chat — and dispatched to a typed handler that returns structured data (a translated query, a resolved URL, a vocabulary record) alongside the natural-language answer. The default model is local Ollama, so there's no third-party LLM dependency unless the customer wires one up explicitly.

The same discipline that the rest of the platform follows applies here too: an AS2 envelope, an S3 file backfill, an AI-translated query — all of them go through the same validation, hashing, Digital-Link normalisation, tenant scoping and role enforcement that direct REST traffic gets. Integrations don't bypass the conformance contract; they're additional ways into it.

## Capabilities by edition

<table>
<thead>
  <tr>
    <th>
      Capability
    </th>
    
    <th>
      OSS
    </th>
    
    <th>
      Business
    </th>
  </tr>
</thead>

<tbody>
  <tr>
    <td>
      AS2 gateway for EDI partners
    </td>
    
    <td>
      <span className="fm-no">
        —
      </span>
    </td>
    
    <td>
      <span className="fm-yes">
        ✓
      </span>
    </td>
  </tr>
  
  <tr>
    <td>
      S3-compatible storage
    </td>
    
    <td>
      <span className="fm-basic">
        ✓ lib
      </span>
    </td>
    
    <td>
      <span className="fm-yes">
        ✓
      </span>
    </td>
  </tr>
  
  <tr>
    <td>
      Azure Blob storage
    </td>
    
    <td>
      <span className="fm-no">
        —
      </span>
    </td>
    
    <td>
      <span className="fm-yes">
        ✓
      </span>
    </td>
  </tr>
  
  <tr>
    <td>
      File upload / download service
    </td>
    
    <td>
      <span className="fm-no">
        —
      </span>
    </td>
    
    <td>
      <span className="fm-yes">
        ✓
      </span>
    </td>
  </tr>
  
  <tr>
    <td>
      AI assistant (Ollama-backed, intent-routed)
    </td>
    
    <td>
      <span className="fm-no">
        —
      </span>
    </td>
    
    <td>
      <span className="fm-yes">
        ✓
      </span>
    </td>
  </tr>
</tbody>
</table>

## See also

- [Architecture → Authentication](/docs/platform-overview/architecture#authentication-one-identity-three-ways-in) — how partner-facing services plug into the platform's identity model.
- [Modules → EPCIS Events](/docs/platform-overview/modules/epcis-events) — what AS2 inbound documents end up being.
- [Roadmap](/docs/platform-overview/roadmap) — additional EDI envelope formats and a retrieval corpus for the AI assistant are pending.