hermes - 💡(How to fix) Fix [Bug]: Model Chooser Conflicts with API Key Resolution and Configuration Management I think It might be related to #12146 [1 participants]

Bug Description

Summary

The model selection interface is causing multiple interconnected issues: an unresolved provider API key error, unexpected configuration modifications when setting local models, hardcoded models overriding user selections, and ambiguity between Ollama API routes and custom local model paths. These problems make reliable model switching difficult and disrupt the workflow for users managing local or hybrid AI deployments.

Current Behavior / Problems Identified

• Receiving "Provider resolver returned an empty API key" error related to OPENROUTER_API_KEY even when using local models.
• The model chooser/switcher UI is cluttered and difficult to navigate.
• Setting a new local model triggers automatic config file modifications that are unstable, unpredictable, or require manual intervention to function correctly.
• Built-in/hardcoded models frequently override user-selected local models without clear indication.
• No intuitive distinction between selecting an Ollama API endpoint versus configuring a custom local model path.

Steps to Reproduce

1. Start a local Ollama server with one or more models available.
2. Open Hermes-webui and navigate to the model selection/switcher interface.
3. Attempt to select or configure the Ollama/local model.
4. Observe the API key error, UI confusion, and hardcoded model overrides.

Expected Behavior

• The provider resolver should correctly detect available local/Ollama models without requiring an OPENROUTER_API_KEY when that provider is not in use.
• The model chooser should provide a clear, organized interface with distinct sections for Ollama API, custom local models, and other providers.
• User-selected models should take precedence over hardcoded defaults, with explicit UI indicators showing which model is active and its source.
• Configuration changes should be transparent, non-destructive, and stored in a way that prevents automatic corruption or unexpected overrides.

Implementation Suggestions

• Decouple the OPENROUTER_API_KEY requirement from local/Ollama model resolution. Only enforce it when explicitly using OpenRouter-hosted models.
• Redesign the model chooser with clear categorization (e.g., "Ollama API", "Custom Local Model", "Other Providers") and collapsible sections to reduce visual clutter.
• Implement a priority system where user-defined models override hardcoded defaults, with visible status indicators showing active selections.
• Use non-destructive config management: store user overrides in a separate configuration file or use a structured format that prevents automatic overwrites during runtime.
• Add validation and context-aware error messaging when API keys are missing but not required for the selected provider route.

I have to have Hermes modify the config every time I switch local models just to be able to use it.

Environment

• Ollama Server.
• OS: [linux]

Additional Context

This issue significantly impacts usability when managing local or hybrid AI workflows. Clarifying the distinction between API-based and local model routing, along with stabilizing the configuration system, would greatly improve reliability and user experience.

Steps to Reproduce

1. Run Hermes model.
2. Choose local ollama model.
3. Run Hermes.
4. New model populated.
5. Send chat.
6. No completion error = "Error: Provider resolver returned an empty API key. Set OPENROUTER_API_KEY or run: hermes setup"

Expected Behavior

Change model = model works.

Actual Behavior

Covered in reproduce.

Affected Component

CLI (interactive chat)

Messaging Platform (if gateway-related)

No response

Debug Report

Report       https://paste.rs/L19MK
  agent.log    https://paste.rs/T3g54
  gateway.log  https://paste.rs/qZNyx

Operating System

Linux Mint 22.3 (Zena), based on Ubuntu Noble.

Python Version

Python 3.12.3

Hermes Version

Hermes Agent v0.11.0 (2026.4.23)

Additional Logs / Traceback (optional)

Root Cause Analysis (optional)

No response

Proposed Fix (optional)

No response

Are you willing to submit a PR for this?

• I'd like to fix this myself and submit a PR