---
title: "Connection architecture"
manual: "AI Assistant"
version: "main"
source: "Developers/Architecture/ConnectionArchitecture.rst"
rendered: "2026-10-03T14:54:45+00:00"
---

# Connection architecture {#connection-architecture}

The connection layer abstracts provider-specific APIs.

## AI connector interface {#ai-connector-interface}

An AI connector is responsible for:

-   chat completions;
-   embeddings;
-   translating provider responses into core DTOs;
-   exposing usage information where available.

AI connection authentication is resolved by the core. Connectors support static API keys and OAuth
2.0 Client Credentials. OAuth token acquisition and renewal are kept outside provider-specific
request code, while each provider maps the resulting bearer credential to its own HTTP API format.

## Vector store connector interface {#vector-store-connector-interface}

A vector store connector is responsible for:

-   creating or validating collections;
-   writing vector documents;
-   searching vectors;
-   deleting documents by source hash;
-   translating provider responses into core DTOs.

## Connection records {#connection-records}

Connection records store runtime configuration and credentials. They are passed
into connectors for every operation. This keeps connector services reusable and
stateless.