Use your own API keys to access frontier models (GPT-4o, Claude, Gemini) through ACAI. All compliance features — guardrails, PII detection, audit logging — apply to passthrough requests.
Passthrough models route requests to external provider APIs (OpenAI, Anthropic, Google) using your API key. ACAI acts as a compliance proxy — enforcing guardrails, logging audit events, and detecting PII before the request leaves your boundary.
Client → ACAI API Server → Guardrails (content safety, PII, injection) → Audit log → Forward to provider (OpenAI / Anthropic / Google) → Response guardrails → Audit log → Client
Store your provider API key in the dashboard. Keys are encrypted with AES-256-GCM at rest and never logged.
Use the same ACAI API key and endpoint. Just specify a passthrough model name:
curl https://api.agilecloud.ai/v1/chat/completions \
-H "Authorization: Bearer $DIRECTAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-4o",
"messages": [
{"role": "user", "content": "Explain HIPAA compliance."}
]
}'ACAI resolves gpt-4o to the OpenAI passthrough config, retrieves your encrypted OpenAI key, and proxies the request.
| Model Alias | Provider | Required Key |
|---|---|---|
| gpt-4o | OpenAI | OpenAI API Key |
| gpt-4o-mini | OpenAI | OpenAI API Key |
| o3 | OpenAI | OpenAI API Key |
| o3-mini | OpenAI | OpenAI API Key |
| claude-4-sonnet | Anthropic | Anthropic API Key |
| claude-3-5-sonnet | Anthropic | Anthropic API Key |
| gemini-2.0-flash | Google AI API Key |
ACAI does not charge per-token for passthrough requests. You pay the provider directly. ACAI's platform fee covers the compliance proxy, guardrails, and audit logging.