AI Providers

Dory uses a pluggable AI provider architecture for AI Chat, SQL generation, chart suggestions, and schema-aware analysis. You can switch providers through environment variables without changing application code.

Supported providers

Provider	`DORY_AI_PROVIDER` value	Notes
OpenAI	`openai`	Uses the official OpenAI API.
OpenAI-compatible	`openai-compatible`	For services exposing an OpenAI-compatible API.
Anthropic	`anthropic`	Claude models via Anthropic's API.
Google	`google`	Gemini models via Google Generative AI.
Qwen	`qwen`	Qwen models through a compatible endpoint.
xAI	`xai`	Grok models via xAI API.

Core variables

Most deployments need the following values:

export DORY_AI_PROVIDER=openai
export DORY_AI_MODEL=gpt-4o-mini
export DORY_AI_API_KEY=your_api_key_here
export DORY_AI_URL=https://api.openai.com/v1

Variable	Required	Description
`DORY_AI_PROVIDER`	Yes	Selects the provider adapter.
`DORY_AI_MODEL`	Yes	Selects the model used by Dory AI features.
`DORY_AI_API_KEY`	Usually	API key or bearer token accepted by the provider.
`DORY_AI_URL`	Provider-dependent	Base URL for OpenAI-compatible providers or custom endpoints.

Do not put provider credentials in frontend code, checked-in .env files, screenshots, or support tickets.

Provider examples

OpenAI

DORY_AI_PROVIDER=openai
DORY_AI_MODEL=gpt-4o-mini
DORY_AI_API_KEY=sk-...
DORY_AI_URL=https://api.openai.com/v1

Use this path when you want the most direct documented setup for Dory AI SQL generation.

OpenAI-compatible

DORY_AI_PROVIDER=openai-compatible
DORY_AI_MODEL=your-model-name
DORY_AI_API_KEY=your_provider_key
DORY_AI_URL=https://your-compatible-endpoint.example.com/v1

Use this path when your organization already standardizes on an OpenAI-compatible API surface.

Anthropic

DORY_AI_PROVIDER=anthropic
DORY_AI_MODEL=claude-3-5-sonnet-latest
DORY_AI_API_KEY=your_anthropic_key

Use Anthropic when your team prefers Claude models for longer context reasoning or policy reasons.

Google

DORY_AI_PROVIDER=google
DORY_AI_MODEL=gemini-1.5-pro
DORY_AI_API_KEY=your_google_key

Use Google when your organization already runs Gemini workflows.

Qwen

DORY_AI_PROVIDER=qwen
DORY_AI_MODEL=qwen-plus
DORY_AI_API_KEY=your_qwen_key

Use Qwen when it is the preferred model family for your language, region, or cost profile.

xAI

DORY_AI_PROVIDER=xai
DORY_AI_MODEL=grok-2-latest
DORY_AI_API_KEY=your_xai_key

Use xAI when Grok models are part of your approved AI stack.

Choosing a provider

Requirement	Recommended direction
Fastest standard setup	Start with OpenAI.
Existing compatible endpoint or proxy	Use OpenAI-compatible.
Long reasoning tasks	Compare Anthropic and OpenAI models.
Existing Google AI usage	Use Google.
Region, language, or cost fit	Compare Qwen and other approved providers.
Internal compliance requirement	Choose the provider already approved by your organization.

Validation checklist

Confirm the provider value matches one of the documented DORY_AI_PROVIDER values.
Confirm the model name exists for that provider.
Confirm the API key has permission to call the selected model.
If using a compatible endpoint, confirm DORY_AI_URL includes the correct base path.
Restart the Dory server after changing provider variables.
Test AI Chat with a small schema question before testing large SQL generation tasks.

Troubleshooting

Symptom	Likely cause	Fix
AI Chat returns authentication errors	Invalid or missing API key.	Rotate the key and update `DORY_AI_API_KEY`.
Model not found	Wrong model name for the selected provider.	Use a model name supported by that provider.
Compatible endpoint fails	Incorrect base URL or path.	Verify the provider's OpenAI-compatible base URL.
SQL quality is inconsistent	Model has limited schema reasoning ability.	Try a stronger model and include clearer table context.
Responses are slow	Provider latency or large schema context.	Use a faster model or narrow the database context.

Limitation

Provider support does not guarantee identical output quality. SQL generation quality will vary by model, schema complexity, and prompt clarity.

On this page