[Bug]: Update of development workspace that has an incomplete Ollama Cloud setup results in an onboarding that is completely blocked / broken

[mpawlow]

OpenRAG Version

0.1.0

Deployment Method

Local development (make dev)

Operating System

Ubuntu 24.04.3 LTS

Python Version

Python 3.13.11

Affected Area

Onboarding (setup wizard, initial configuration)

Bug Description

Update of development workspace that has an incomplete Olama Cloud setup results in an onboarding that is completely blocked / broken

Steps to Reproduce

• (1) Rebuild all images via make build-be
• (2) Run make dev-local-cpu, make docling, make backend, make frontend
• (3) Go to OpenRAG Frontend app URL
  • e.g. http://openrag1.fyre.ibm.com:3000/
  • Setup wizard loads and detects previous OpenAI model provider configuration
  • See Screenshot 1
• (4) Click the "Complete" button
  • BUG: "OpenAI API error: The model nomic-embed-text:latest does not exist or you do not have access to it. (type: invalid_request_error) (code: model_not_found)"
  • The nomic-embed-text:latest model is not applicable to OpenAI - it is a Ollama model
  • BUG: Completely blocked. Unable to proceed
• (5) Click the "Ollama" card
  • Previous incomplete / broken Ollama Cloud model configuration is loaded
  • BUG: Unable to edit or remove the Ollama Cloud configuration
  • See Screenshot 3

Expected Behavior

• Should not mix up models between model providers
• Should be able to override or clear pre-existing Ollama model provider configuration
• Should be able to override or clear pre-existing OpenAI model provider configuration

Actual Behavior

As described above

Relevant Logs

{
    "error": "Ollama API error: {\"error\":\"path \\\"/api/embeddings\\\" not found\"}\n"
}

Screenshots

Screenshot 1

Screenshot 2

Screenshot 3

Additional Context

• Related: [Feat]: Settings UX should let you remove a model provider #833

Checklist

• I have searched existing issues to ensure this bug hasn't been reported before.
• I have provided all the requested information.

[mpawlow]

Work-around

Here are several ways to reset the OpenAI/Ollama embedding model configuration:

Option 1: API Rollback (Recommended)

Call the rollback endpoint which fully resets onboarding and clears model config:

  curl -X POST http://localhost:8000/onboarding/rollback

This will:

• Cancel active tasks
• Delete ingested documents from OpenSearch
• Clear embedding_model and reset embedding_provider to "openai"
• Set edited: false so you can re-onboard

Option 2: Manually Edit the YAML Config File

The config file is at config/config.yaml. You can edit it directly:

Stop the backend first, then edit:

  nano config/config.yaml

Reset these fields:

  providers:
    openai:
      api_key: ""
      configured: false
    ollama:
      endpoint: ""
      configured: false

  knowledge:
    embedding_model: ""
    embedding_provider: "openai"

  onboarding:
    current_step: 0
    assistant_message: null
    selected_nudge: null
    card_steps: null
    upload_steps: null
    openrag_docs_filter_id: null
    user_doc_filter_id: null

  edited: false

Then restart the backend.

Option 3: Delete the Config File Entirely

  rm config/config.yaml

The ConfigManager will recreate it with defaults from environment variables on next startup.

Option 4: Factory Reset (Nuclear Option)

  make factory-reset

This stops everything, removes Docker volumes, data directories, and images - a complete reset.