Models

The Models page is where you connect the large language model providers that power every run: OpenAI, Anthropic, Google, OpenRouter, Azure OpenAI and others. You bring your own key (BYOK), so model usage is billed by the provider to your own account, and the page shows what each provider is costing you.

Connect at least one provider before you build a worker: every agent, team and AI employee needs a model to run. Come back here to rotate a key, add Azure deployments, give an AI employee an image or video model, or find out which model and which part of the product is driving your spend.

How models reach a worker

Connect and manage models step by step

Open Models under Configure in the left sidebar. The Models page is shown to organization owners and admins, who are the only roles that can connect or disconnect an organization's providers.

1. Read the overview

   The top of the page summarises model usage for the period you pick with 24h,7d, 30d or QTD (quarter to date):TOTAL COST with the daily average, TOTAL TOKENS, AVG LATENCY with its p95, and PROVIDERS CONNECTED. Below it, Connected providers shows one card per provider you have connected, and Available providers lists the ones you have not. Use Search providers… and the type filter to narrow the list.

   app.turtleaicoworker.com/models

   

2. Connect a provider with your own key

   Under Available providers, each card shows how many models the provider offers, a short description, and how it connects (API key for most). ClickConnect.

   app.turtleaicoworker.com/models

   

   In the Connect dialog, paste the key into API key. TheGet your … API key link opens the provider's own key page if you need to create one. A checklist below it shows what the connection gives you: the provider's models across your agents and teams, token usage and per-model costs, latency and error monitoring. ClickUpdate API Key to save. The key is encrypted at rest and in transit and is never written to logs.

   

3. Rotate or remove a key

   On a connected provider's card, click Update key (also in the card's… menu as Update Key). The dialog shows the active key masked, for example sk-...****. Paste the new key and click Update API Key: every request uses the new key straight away. To remove the provider, click Disconnect in the same dialog or in the … menu.

   

4. Connect Azure OpenAI

   Azure OpenAI works differently, because on Azure you create the models yourself, as deployments in your own Azure resource. Click Connect on Azure OpenAI and fill in the Azure endpoint (for example https://your-resource.openai.azure.com/), the API version and the API key, then click Connect.

   The Add your first model dialog opens straight after. For each deployment, enter theDeployment name exactly as it appears in Azure, and the Base model name it was created from, such as gpt-5.4 or gpt-image-2. UseAdd another model for more, then click Deploy models. You can also choose Skip for now and add deployments later.

   

5. Understand how an Azure deployment is classified

   The base model name decides what a deployment is, never the deployment name you chose. A deployment of an image or video base model is treated as a media model; every other base model is a chat model that workers can reason with. So you never have to say which kind it is, and renaming a deployment does not change it.

   • Image models: gpt-image-1, gpt-image-1-mini, gpt-image-1.5, gpt-image-2, dall-e-2, dall-e-3.
   • Video models: sora-2, sora-2-pro, sora-2-pro-high-res.

   A dated snapshot name, such as gpt-image-1.5-2025-12-16, is recognised as its base model. Each deployment is checked with Azure when you add it. A chat deployment gets a short test call, including a tool call, because workers drive their tools through tool calls. An image or video deployment is not given a test render, which Azure would bill you for: it is checked against the list of deployments on your resource instead.

6. Open a provider's usage view

   Click View models on a connected card. The provider page has a period selector (Last 7 days, Last 2 weeks, Last month, Last 3 months, Last 6 months) and shows:

   • TOTAL SPEND, REQUESTS, TOKENS (input and output), SUCCESS rate and AVG LATENCY, each with the change against the previous period;
   • CONNECTION (who connected it and whether it is org-wide), MODELS AVAILABLE and SHARE OF TOTAL LLM SPEND;
   • Spend trend by day and Cost-share by model.

   app.turtleaicoworker.com/models

   

   Further down, Usage by module splits spend between teams, AI employees and agents, andEffective blended rate gives the cost per 1K tokens with the cheapest and priciest model. The MODELS table lists each model with its requests, tokens, cost, cost per 1K tokens, success rate, latency, when it was last used, and its TOP DRIVER, the worker that used it most. Click a column to sort. For Azure OpenAI, Add model here opens the same deployment dialog as in the previous step.

   app.turtleaicoworker.com/models

   

   Image and video renders count toward the provider that produced them, with the model named after its deployment on Azure. They are billed per render rather than per token, so they appear in spend and cost share even when their token count is zero.

7. Choose a model for each worker

   Each worker runs on the model you set on it, from any provider you have connected:

   • Sequential agent: in the agent's settings, pick the LLM Provider and then the LLM Model.
   • Agentic team: in the team settings, pick the LLM Provider and LLM Model. The team runs on the model you set here.
   • AI employee: in Behavior Settings, use LLM Configuration to pick the provider and model.

   Only chat models appear in these pickers. If none appear, no provider is connected yet: an owner or admin needs to connect one on the Models page first.

8. Give an AI employee image and video generation

   An AI employee can make images and videos as part of its work. In the employee's Behavior Settings, the Image Generation and Video Generation sections each have a Provider and a Model picker. Pick a model and clickSave; the section then confirms that the employee can generate images or videos. Both are optional, and Disconnect removes the model again.

   Only connected providers that offer image or video models are listed, including Azure deployments of the image and video base models above. Save a new employee first; the media sections become usable once it exists.

Reference

Provider connection

Connected provider card

Where each worker's model is set

Related

• Tools & Integrations: connect the services workers act in.
• Custom Tools & MCP servers: add your own APIs and MCP servers.
• Sequential Agents, Agentic Teams, AI Employees: where each worker's model is chosen.