Connect OpenAI, Anthropic, Gemini, Mistral, Groq or OpenRouter to Xcellerate AIG

Connecting OpenAI, Anthropic, Google Gemini, Mistral, Groq or OpenRouter to Xcellerate AIG gives your team one governed entry point to hosted AI models. Your applications call AIG with their own virtual key and never hold the vendor key. This guide is for the admin who sets up AIG for an MSP, SME or service business.

TL;DR

• Setup is the same for all six vendors: pick the type, give it a name, paste one API key.
• A new provider starts disabled, so you can test it with one small request first.
• After enabling it, sync the model catalogue and issue a virtual key for each application.

Before you start

• An account with the vendor and a standard API key: an OpenAI key, an Anthropic console key, a Google AI Studio (Gemini API) key, or a Mistral, Groq or OpenRouter key.
• The Admin role in AIG, or a custom role with permission to create and edit providers.
• Outbound HTTPS from the AIG host to the vendor. The default endpoints are api.openai.com/v1, api.anthropic.com/v1, generativelanguage.googleapis.com/v1beta, api.mistral.ai/v1, api.groq.com/openai/v1, openrouter.ai/api/v1. If traffic must go through a proxy, set it under General settings → Network egress.
• No callback URLs or webhooks are needed.

Set up the provider

1. Start the wizard

In the sidebar, go to Connect → Providers and click Connect a provider. If you prefer the full form, choose Set up manually or Add provider.

2. Choose the vendor and name it

Under Provider type, pick OpenAI, Anthropic, Google Gemini, Mistral, Groq or OpenRouter. Give it a Display name people will recognise, for example "OpenAI production". Leave Base URL empty to use the vendor's default endpoint.

3. Paste the API key

Paste the key into API key. AIG saves the provider as disabled, so no traffic reaches it yet.

4. Try it

Send one small, real request with Try it. It costs a few tokens. You can also choose Continue anyway; the provider then stays disabled. Later, Test connection runs the same check.

5. Enable, sync and issue a key

Click Enable provider. Then load the model catalogue with Models → Sync, and create a key for your application under Governance → Virtual keys → New virtual key.

Optional: more keys, narrower scope

• Use Add key to put several vendor keys on one provider, and give each a Weight to spread traffic.
• Limit a key to certain models with Allowed models.
• Under What this provider serves, tick only what you need.

What happens after you connect

Your applications address a model as <provider-name>/<model>, for example openai/gpt-4o-mini, or by a bare model name that AIG resolves through the catalogue. AIG translates each request into the vendor's format, so one OpenAI-style request can reach OpenAI, Anthropic or Gemini.

The catalogue sync pulls model names, context windows and prices from a published datasheet, on a schedule (at most once an hour) or on demand. If you have negotiated prices, you can override them.

What each type serves:

• OpenAI: chat, embeddings, images, speech, transcription, rerank
• Anthropic: chat
• Google Gemini: chat, embeddings
• Mistral: chat, embeddings, transcription
• Groq: chat, transcription, speech
• OpenRouter: chat, embeddings, images

Good to know

• Max retries is 0 by default. Switch retries on deliberately.
• With several keys, AIG rotates between them automatically. A rejected key is marked Dead or Suspect; a rate-limited key shows Cooling down and comes back by itself.
• Disabling a provider stops traffic from the next request and keeps its keys. Deleting a provider also deletes its keys.
• Streamed OpenAI-style requests are only priced when the caller sends stream_options: {"include_usage": true}.
• API keys are write-only: after saving, you only see a masked hint such as sk-…4f2a, and vendor keys never appear in error messages.
• Without a licence, AIG runs in free mode: 5 virtual keys with 5 requests per key per day. A licence removes the cap.

Troubleshooting

• "No enabled key to test with." Add a key or enable the existing one, then test again.
• "No model to try. Sync the catalog first…" Run Models → Sync and try again.
• <model> did not answer (<ms> ms): <message> The message comes from the vendor. Check the key and whether your account has access to that model.
• No healthy API key is available for provider '…' (503) or Every API key configured for '…' failed for this request. (502): check the key status (Healthy, Dead, Cooling down, Suspect), fix the key, then click Clear health flags. No enabled provider is configured for '…' (400) means no enabled provider matches that name.

Talk to us about AIG

Want to see how AIG fits your MSP or service business? Talk to us about AIG. Read more about every provider in one catalogue and virtual keys.