Provider Setup

Route requests to Gemini, OpenAI, Ollama, or any OpenAI-compatible endpoint by setting two environment variables.

PromptShield speaks OpenAI's chat completions API on the inbound side and translates to whatever your upstream provider expects. Adding a new provider is one .env change.

Set PROMPTSHIELD_PROVIDER and your API key. That is it.

Gemini

PROMPTSHIELD_PROVIDER=gemini
GEMINI_API_KEY=your-gemini-key

# override the model (optional):
# PROMPTSHIELD_GEMINI_MODEL=gemini-2.0-flash

Default model: gemini-2.0-flash

OpenAI

PROMPTSHIELD_PROVIDER=openai
OPENAI_API_KEY=your-openai-key

# override the model (optional):
# PROMPTSHIELD_OPENAI_MODEL=gpt-4o-mini

Default model: gpt-4o-mini

Ollama / self-hosted

No API key needed for local models.

PROMPTSHIELD_PROVIDER=selfhosted
PROMPTSHIELD_UPSTREAM_URL=http://localhost:11434/v1
PROMPTSHIELD_SELFHOSTED_MODEL=llama3.1:8b

Default model: llama3.1:8b

Any OpenAI-compatible endpoint

Together, Groq, Fireworks, Anyscale. If it speaks OpenAI, PromptShield speaks to it.

PROMPTSHIELD_PROVIDER=openai-compatible
PROMPTSHIELD_UPSTREAM_URL=https://api.together.xyz/v1
PROMPTSHIELD_UPSTREAM_API_KEY=your-key
PROMPTSHIELD_MODEL=mistralai/Mixtral-8x7B-Instruct-v0.1

Model resolution order

PROMPTSHIELD_MODEL (global override) → provider-specific env var (PROMPTSHIELD_GEMINI_MODEL, etc.) → built-in default.

The model field in the request body is always ignored. Model selection is a server-side concern in PromptShield.