Local OpenAI-Compatible Providers Missing from Hermes Desktop Configuration

Bug tracker — July 11, 2026. Hermes bug #62213 exposes a Desktop configuration gap: after a clean macOS install, the UI lists hosted providers but offers no Custom or local endpoint option for llama.cpp, LM Studio, Ollama, or other OpenAI-compatible servers. CLI/config naming is also inconsistent (custom, local, custom:local), leaving local-model users unable to configure providers through the primary interface.

The Failure in One Sentence

Hermes Desktop claims to support local OpenAI-compatible providers, but a clean installation may present no UI path to configure them. Users must guess inconsistent CLI/config names or edit configuration outside the Desktop. The result is a local-model feature that exists in the architecture but is unreachable through the interface most users are expected to use.

The Configuration Mismatch

SurfaceObserved BehaviorUser Impact
Desktop provider pickerHosted providers only; no Custom/local endpointCannot add llama.cpp, LM Studio, Ollama, or similar servers
CLI/config namingcustom, local, and custom:local are inconsistentUsers cannot infer the supported configuration key
Documentation pathUI and config model disagreeSetup requires trial-and-error or manual file edits

Why Desktop Access Matters

For users evaluating Hermes as a local-model platform, Desktop is not a secondary surface. It is the discovery and configuration path. Hiding local providers from that path creates a misleading product experience: local backends may work after manual intervention, but the normal onboarding flow suggests they do not exist. This especially affects users who are not comfortable editing provider configuration by hand.

Recommended Fix Path

  1. Add a first-class Custom/OpenAI-compatible provider form to Desktop with endpoint, model, and authentication fields.
  2. Normalize provider identifiers across Desktop, CLI, and config files; accept aliases only with explicit migration guidance.
  3. Validate the endpoint and show a clear connection test before saving.
  4. Add clean-install regression coverage for Ollama, LM Studio, llama.cpp, and generic OpenAI-compatible URLs.

Why This Matters for Review-Site Readers

A local provider that cannot be configured through the supported Desktop workflow is a practical availability issue, not merely a UX omission. Comparison shoppers should distinguish between nominal backend compatibility and an end-to-end path that lets ordinary users discover, configure, test, and switch to that backend.

Case Summary

Source URL: https://github.com/NousResearch/hermes-agent/issues/62213