Bring your own model: Concierge docs

Out of the box, every agent runs on Chargezen Managed models, paid from your Concierge credits. That's the right choice for most businesses. If your company already has an agreement, credits or data terms with an AI provider, you can run your agents on your own account instead. Calls on your own key are billed by your provider and never charged to your Chargezen credits.

Providers you can connect

OpenAI: API key, plus organization and project IDs if you use them.
Anthropic: API key, for Claude models.
Google Gemini: a Google AI Studio API key.
xAI: API key, for Grok models.
Fireworks: API key (and account ID if you have one), for open models such as DeepSeek, Kimi, Qwen and GPT-OSS.
AWS Bedrock: access key ID, secret access key and region (and a session token if you use one), to run models such as Claude and Amazon Nova under your AWS account.
Deepgram: API key, for speech-to-text and text-to-speech on voice calls.

Each provider offers a short list of models chosen to work well with Concierge's tools. The list for each provider shows up in the agent's model menu once you connect it.

Connect your provider

Create a key with your provider

In your provider's console, create an API key in the account or project you want billed. Keep it private.

Open Integrations

In the Chargezen admin, go to Deploy > Integrations and open the AI category.

Add the key

Open the provider, go to Setup, fill in the fields and click Connect.

Choose the model for an agent

Open the agent

Go to Deploy > Agents and click Configure on the agent.

Pick the default model

Under Models, open Default model. Your connected providers appear as "Provider · Model", next to Chargezen Managed. Pick one.

Optional: request options

A model on your own account shows a request options box for JSON settings your provider accepts. Leave it as is unless you know you need a setting. For Anthropic models used on voice, give a positive max_tokens value.

Optional: overrides

Open Model overrides and plan revisions to use a different model for specific parts of the agent's work. Each override can stay on Use default.

Save

Click Save Agent. The agent's next piece of work uses the new model.

You can mix: one agent on your own OpenAI account, another on Chargezen Managed. Tools, approvals, memory and channels work the same on any model.

Voice on your own Deepgram account

By default Chargezen runs speech recognition and the agent's voice. After you connect Deepgram, the agent's Voice calls settings show Speech-to-text and Text-to-speech menus under Language and speech, where you choose whose account runs each one.

Other AI features

A few AI features outside your agents, such as AI SMS drafts and AI reply suggestions, are listed in Settings > Credits under Features that use credits. When your workspace has a connected OpenAI, Anthropic, Fireworks or AWS Bedrock account, they run on it and show "Runs on your connected model provider."

Not the same as the ChatGPT or Claude connector

This page sets the model that powers your agents. Using your Chargezen workspace from the ChatGPT or Claude app is a separate connection; see ChatGPT, Claude and MCP.

Troubleshooting and FAQ

Check the provider shows as connected on the Integrations page, then reopen the agent. Only providers that can run agents appear in the menu.

The agent started failing after I switched models

Check the key is still valid and your provider account has billing set up, then reconnect it on the Integrations page. To get going again right away, switch Default model back to Chargezen Managed.

Do I still use credits with my own key?

Not for model calls on your key. Speech on Chargezen managed speech-to-text and text-to-speech still uses credits, and phone and text charges are billed as used.

Which model should I pick?

Start with Chargezen Managed. Switch when you have a reason: a provider agreement, data terms your company requires, or a model your team has already tested.

Guides

On the product page

Bring your own model