SuperagentProviders
Supported LLM providers and configuration
Providers
Safety Agent works with any language model. Use the provider/model format when specifying models.
Supported Providers
| Provider | Model Format | Required Env Variables |
|---|---|---|
| Superagent | superagent/{model} | None (default for guard) |
| Anthropic | anthropic/{model} | ANTHROPIC_API_KEY |
| AWS Bedrock | bedrock/{model} | AWS_BEDROCK_API_KEYAWS_BEDROCK_REGION (optional) |
| Fireworks | fireworks/{model} | FIREWORKS_API_KEY |
google/{model} | GOOGLE_API_KEY | |
| Groq | groq/{model} | GROQ_API_KEY |
| OpenAI | openai/{model} | OPENAI_API_KEY |
| OpenRouter | openrouter/{provider}/{model} | OPENROUTER_API_KEY |
| Vercel AI Gateway | vercel/{provider}/{model} | AI_GATEWAY_API_KEY |
Environment Setup
The Superagent guard model is used by default and requires no API keys. Set the appropriate API key environment variable only if you want to use a different provider or need the redact() method (which requires a model):
# Superagent (optional - for usage tracking only)
export SUPERAGENT_API_KEY=your-key
# OpenAI
export OPENAI_API_KEY=sk-...
# Anthropic
export ANTHROPIC_API_KEY=sk-ant-...
# Google
export GOOGLE_API_KEY=...
# Groq
export GROQ_API_KEY=gsk_...
# Fireworks
export FIREWORKS_API_KEY=...
# AWS Bedrock
export AWS_BEDROCK_API_KEY=...
export AWS_BEDROCK_REGION=us-east-1 # optional
# OpenRouter
export OPENROUTER_API_KEY=...
# Vercel AI Gateway
export AI_GATEWAY_API_KEY=...Usage Examples
import { createClient } from "@superagent-ai/safety-agent";
const client = createClient();
// Superagent (default - no API key required)
await client.guard({
input: "user message"
// model defaults to superagent/guard-0.6b
});
// Or specify Superagent model explicitly
await client.guard({
input: "user message",
model: "superagent/guard-0.6b"
});
// OpenAI
await client.guard({
input: "user message",
model: "openai/gpt-4o-mini"
});
// Anthropic
await client.guard({
input: "user message",
model: "anthropic/claude-3-5-sonnet-20241022"
});
// Google
await client.guard({
input: "user message",
model: "google/gemini-1.5-pro"
});
// Groq
await client.guard({
input: "user message",
model: "groq/llama-3.1-70b-versatile"
});
// OpenRouter (nested provider/model)
await client.guard({
input: "user message",
model: "openrouter/anthropic/claude-3-5-sonnet"
});Vision-Capable Models
For image analysis, use a vision-capable model:
| Provider | Vision Models |
|---|---|
| OpenAI | gpt-4o, gpt-4o-mini, gpt-4-turbo, gpt-4.1 |
| Anthropic | claude-3-*, claude-sonnet-4-*, claude-opus-4-*, claude-haiku-4-* |
gemini-* |
Other providers (Fireworks, Groq, OpenRouter, Vercel, Bedrock) currently support text-only analysis.
Choosing a Model
Default: Superagent Guard Model
The default superagent/guard-0.6b model is recommended for most guard use cases:
- No API keys required - works out of the box
- Low latency - optimized for fast classification
- Proven accuracy - purpose-trained for safety classification
- No cost - free to use
Other Models
Consider these factors when selecting a different model:
- Latency: Smaller models like
gpt-4o-miniorclaude-3-haikuare faster - Accuracy: Larger models like
gpt-4oorclaude-3-5-sonnetmay catch more edge cases - Cost: Varies significantly by provider and model size
- Compliance: Some providers offer data residency or compliance certifications
For guard use cases where you need a different provider, openai/gpt-4o-mini or anthropic/claude-3-haiku-20240307 provide a good balance of speed, accuracy, and cost.