Aller au contenu

Models

Ce contenu n’est pas encore disponible dans votre langue.

You decide which model the agent thinks with. Manage models on the AI providers page: open the account menu from your avatar and choose Providers.

Models marked built-in work without any setup. Today they run on Cloudflare Workers AI:

  • Kimi K2.7 Code
  • GLM 5.2
  • GLM 5.3 Flash
  • DeepSeek V4 Pro

Choose Add provider and pick one:

ProviderWhat you enter
Anthropic (such as Claude Opus 5, Claude Sonnet 5, Claude Haiku 4.5)Model, API Token
OpenAI (such as the GPT 5.6 family)Model, API Token
GoogleModel, API Token
OllamaModel ID, API URL (a reachable address for your server, like the one behind http://localhost:11434)
Cloudflare Workers AIModel ID, Cloudflare Account ID, API Token

You can also set a Display Name shown in the interface. Advanced Settings lets you override the default API endpoint, useful for proxies such as Cloudflare AI Gateway.

After adding a model, run the test to check the key and model work.

Some small tasks, like naming a conversation, don’t need the strongest model. Click a row in the model list to make it the Quick model for those tasks; click again to clear it. Pick something fast and inexpensive.

Switch the model for the current conversation from the workspace input. Use a stronger model for long jobs and hard reasoning, and a faster one for simple rewrites and formatting.