Runtime Managed Connector Providers

Wiki source: runtime-managed connector provider dossier

Runtime Managed Connector Providers

Administrators can manage custom OpenAI-compatible provider definitions and built-in availability from /connectors/providers. Built-ins expose safe status-only controls and retain their existing adapters. Existing connectors are linked during migration without rewriting credentials or model history. Custom model identity is carried from connector discovery through routing.

• Requirements, scope and status
• Architecture decision ADR-157
• API reference
• QA evidence and open release lanes

Update 2026-10-04: Hugging Face, Pollinations, NVIDIA NIM and AI Horde run as data on a definition (base URL, list format, chat path, key header), not as adapters; recipes are in the provider catalog. Custom providers show their own name everywhere. Open: RBAC, device matrix, accessibility/performance and GitHub CI lanes; production was not reachable from the work machine, so the production fix is the same chat-path setting applied there after deploy.