Skip to main content

AI Provider Configuration

Parthenon's AI assistant, Abby, supports multiple AI backends. The AI Provider Configuration page lets administrators choose which provider powers Abby, configure credentials, and verify connectivity. Navigate to Admin > System > AI Providers to manage providers.

Single-Active Provider Architecture​

Only one AI provider can be active at any given time. The active provider is highlighted at the top of the page in a status banner showing the provider name, selected model, and an "Active" badge. When you set a different provider as active, the previous one is automatically deactivated within a database transaction.

Supported Providers​

Parthenon ships with 8 pre-configured AI providers spanning local, US-hosted, EU-hosted, and China-hosted options:

ProviderRegionModelsAuthentication
OllamaLocalMedGemma 1.5 4B, Llama 3.2, Gemma 3, MistralBase URL (no API key)
AnthropicUSClaude Opus 4, Claude Sonnet 4, Claude HaikuAPI key
OpenAIUSGPT-4o, GPT-4o Mini, o3-miniAPI key
Google GeminiUSGemini 2.5 Pro, Gemini 2.0 Flash, Gemini 1.5 ProAPI key
Mistral AIEUMistral Large, Mistral MediumAPI key
DeepSeekChinaDeepSeek Chat, DeepSeek ReasonerAPI key
QwenChinaQwen Max, Qwen Plus, Qwen TurboAPI key
MoonshotChinaMoonshot v1 128KAPI key

Each provider card displays a region badge indicating where the inference requests are processed. This is important for data governance and regulatory compliance in clinical research environments.

Provider Card Layout​

Each provider is displayed as an expandable card with:

  • Provider name and current model
  • Active badge (shown only for the currently active provider)
  • Region badge (Local, US, EU, or China)
  • Enable/Disable toggle -- controls whether the provider is available for activation
  • Expand/Collapse chevron -- click the card header to reveal configuration options

Configuring a Provider​

  1. Click a provider card to expand it.
  2. Select a model from the dropdown menu. Each provider offers a curated list of supported models.
  3. Enter credentials:
    • For cloud providers: enter the API Key in the masked input field. Click the eye icon to toggle visibility.
    • For Ollama: enter the Base URL (defaults to http://localhost:11434).
  4. Click Save to persist changes. All API keys are stored encrypted in the database using Laravel's encrypted:array cast.

Testing a Connection​

Before activating a provider, verify that it can respond to requests:

  1. Expand the provider card.
  2. Click Test Connection.
  3. A result banner appears below the configuration fields:
    • Green with a checkmark -- the provider responded successfully.
    • Red with an X -- the connection failed. Review the error message for details (invalid key, unreachable endpoint, model not found, etc.).

Setting the Active Provider​

  1. Expand the desired provider card.
  2. Click Set as Active.
  3. The provider immediately becomes the backend for all AI-assisted features across the platform.

If a provider is already active, the button reads "Currently Active" and is disabled.

Enable/Disable Toggle​

Each provider has an independent enable/disable toggle separate from the active state. Disabling a provider hides it from the active selection without removing its configuration. This is useful for temporarily removing a provider from consideration (for example, if an API key has expired) without losing its saved settings.

Ollama for air-gapped environments

For healthcare organizations operating in air-gapped or highly regulated environments, Ollama runs entirely on-premises. The default configuration uses MedGemma 1.5 4B, a medical domain-tuned model. No patient data or queries leave your infrastructure. Set the base URL to point to your internal Ollama instance.

API key security

API keys are stored encrypted at rest, but exercise caution when entering them. Never share keys across environments (development vs. production). Rotate keys immediately if you suspect they have been exposed. Parthenon does not display stored API keys in full -- they are masked in the UI after saving.

Unsaved Changes​

When you modify the model selection, API key, or base URL for a provider, a Save button appears in the action row. Changes are not persisted until you explicitly click Save. If you collapse the card or navigate away without saving, changes are lost. The Save button is hidden when no modifications have been made.

Configuration Workflow​

A recommended workflow for configuring a new provider:

  1. Enable the provider using the toggle switch.
  2. Expand the card and select the desired model.
  3. Enter credentials (API key or base URL).
  4. Save the configuration.
  5. Test the connection to verify the credentials and model availability.
  6. Set as Active once the test succeeds.

Following this order ensures you do not activate a provider before confirming it works correctly.

Data Governance Considerations​

When selecting a provider for clinical research workloads, consider:

  • Local (Ollama) -- no data leaves your infrastructure; suitable for PHI-adjacent queries.
  • US providers -- subject to US data protection laws; review your BAA coverage.
  • EU providers -- subject to GDPR; may be required for EU-based institutions.
  • China providers -- subject to Chinese data regulations; evaluate compliance requirements before use.