Provider Setup

Route requests to Gemini, OpenAI, Anthropic, Ollama, or any OpenAI-compatible endpoint.

Set PROMPTSHIELD_PROVIDER and your API key.

Gemini

PROMPTSHIELD_PROVIDER=gemini
GEMINI_API_KEY=your-gemini-key
PROMPTSHIELD_GEMINI_MODEL=gemini-2.0-flash

OpenAI

PROMPTSHIELD_PROVIDER=openai
OPENAI_API_KEY=your-openai-key
PROMPTSHIELD_OPENAI_MODEL=gpt-4o-mini

Anthropic

PROMPTSHIELD_PROVIDER=anthropic
ANTHROPIC_API_KEY=sk-ant-api03-xxxx
PROMPTSHIELD_ANTHROPIC_MODEL=claude-sonnet-4-5

Ollama / self-hosted

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

No API key needed for local models.

Any OpenAI-compatible endpoint

Together, Groq, Fireworks, Anyscale, or anything that speaks /v1/chat/completions.

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

Multi-provider mode

Route each request to the right backend based on the model name in the request body. Set PROMPTSHIELD_PROVIDERS to a comma-separated list. The first entry is the fallback for unrecognised models.

PROMPTSHIELD_PROVIDERS=anthropic,openai,gemini
ANTHROPIC_API_KEY=sk-ant-...
OPENAI_API_KEY=sk-...
GEMINI_API_KEY=...

Built-in routing rules:

Model prefix	Provider
`claude-*`	Anthropic
`gpt-`, `o1`, `o3`, `o4`	OpenAI
`gemini-*`	Gemini

Override or add routes with PROMPTSHIELD_MODEL_ROUTES:

PROMPTSHIELD_MODEL_ROUTES=llama3=selfhosted,mistral=openai-compatible

Model selection

No default model. Set one via env var or pass model in every request.

The model field in the request body takes priority. In multi-provider mode it also determines which backend handles the request.

Resolution order:

model in the request body
PROMPTSHIELD_MODEL : global override
Provider-specific env var (PROMPTSHIELD_GEMINI_MODEL, PROMPTSHIELD_OPENAI_MODEL, etc.)