Models and AI Providers

01Where AI Providers is#

AI Providers is its own section in Preferences (Edit ▸ Preferences, right after System). The gear at the right end of the chat's header opens it too. It lists, in order: Angelo team (where you sign in to a team), Get started on your first visit, Coding agents (the accounts the chat uses), Generative AI (keys for images and 3D), Agent permissions, AI Enhance and External AI apps (MCP). Add connection at the top adds one.

You don't have to go there to start: an empty chat offers the same ways to connect in place of the message box. Angelo can tell you it found a sign-in on your computer, but it connects only when you click.

02Claude and ChatGPT subscriptions#

Angelo never sees or copies your password. If a sign-in doesn't finish, nothing is left behind: click Connect again. When a chat is connected to a subscription you already have in Angelo, the button reads Use Claude or Use ChatGPT and switches the chat to it.

03API keys#

Choose Use an API key in an empty chat, or Add connection in AI Providers. Pick what the key is for, the provider, paste the key and click Connect; Angelo checks it and says when it is ready. Usage is billed to that provider account.

Remember in Keychain on a Mac, or Remember in Credential Manager on Windows, is on by default and keeps the key in your computer's own store. Turn it off and the key stays in memory only; you enter it again after a restart. An OpenAI coding key is kept by Angelo's Codex in its own store instead. Each connection's row says where its key is.

To change a key, choose Replace key… in the row's ⋯ menu. Remove connection… erases the saved key and stops that connection's chats. It doesn't sign you out of the provider or touch your account there.

04Choosing a model#

The button beside Send shows the chat's model and effort. Click it, then click the model's name to open Select model: a search box, a Default row, and every model from every connected coding account, grouped by provider. Picking a model from another account switches the chat to that account; the conversation stays.

Every release is tested on Claude Opus 5.5, GPT-6 Astra and GPT-6.1 Sol. The list shows whatever your account offers, so you may see more.

Reasoning effort

The slider sets how long the model thinks before it acts, from the efforts that model offers (for example Low, Medium, High). A new chat starts at Low. More effort makes each reply slower and costlier, and in our tests rarely better; raise it only when a request keeps going wrong. The reset button returns to the model's own default.

Helper models

The agent may consult a faster model on the same account, for example Claude Haiku, without asking; each consultation is shown in the chat and uses your quota. Turn this off with Run fast helper models without asking in Preferences ▸ AI Providers ▸ Agent permissions. Other accounts or providers always ask.

Fast

Models with a paid priority tier show a lightning bolt. Turning it on asks first: Fast can cost more, and on Claude it uses paid usage credits rather than your plan's included usage. Fast turns off when the chat changes account.

05Limits and errors#

When a provider says you've reached a usage limit, the chat shows Usage limit reached with the reset time when the provider gives one, and Switch connection to carry on with another account. The provider's own message is under Details. Other problems say what to do in plain words: Sign-in expired (sign in again), Not enough credits, Conversation too long (start a new chat), Can't reach the provider (check your connection).

Your requests go from your computer straight to the provider you chose; Angelo's servers are not in between, unless you use a team connection (Teams).