Model providers

Choose a provider guide for the endpoint and credentials you use:

• Local and OpenAI-compatible models: Ollama, vLLM, LM Studio and llama.cpp, gateway routing for hosted flows, private network access, and Gemini/Vertex verification limits.
• Amazon Bedrock: AWS credentials and regional model access.
• Azure OpenAI: deployment names, API keys and Entra ID.
• Alibaba Cloud Model Studio: regional endpoints and model discovery.

After registration, select the model on a flow and check which gateway URL the agent calls.

Rotate a provider key

Replace a model's API key, or its Vault or OpenBao reference, without editing each model that uses it:

curl -X POST "$PRELOOP_URL/api/v1/ai-models/<model_id>/credentials/rotate" \
  -H "Authorization: Bearer $PRELOOP_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"api_key": "<new key>"}'

Send external_ref instead of api_key for a Vault or OpenBao secret. The new secret is verified before it is saved; if verification fails the old one stays in force. Every model that shares the secret uses the new one on its next request. Models show credentials_version and credentials_rotated_at, and the audit log records the rotation without the key. Subscription logins (Claude Code, Codex) rotate through the provider's own refresh instead.