AI providers

Context Code is not tied to a single provider. Connect the ones you want and pick, chat by chat (even message by message), which model to work with.

How each one connects

In the model settings panel the providers come grouped:

Group	How it connects	Who
Sign-in (OAuth)	Your browser opens and you authorize with your account. No keys to copy.	Anthropic Claude, OpenAI GPT, GitHub Copilot
Developer access (API key)	You paste a key you get from the provider’s website.	Gemini API Key, xKiro, OpenRouter, DeepSeek, xAI (Grok), Zhipu AI (ZAI), MiniMax, Alibaba (Qwen), Kimi (Moonshot), Groq, Cerebras, Nvidia NIM, HuggingFace, Amazon Bedrock, Google Vertex AI, Azure AI Foundry, Meta AI, Xiaomi MiMo and more
Local	No account needed.	Ollama, LM Studio and Context Forge, the app’s own embedded engine

The “Available Engines & AI” dropdown: every provider with its status — Connected or Unauthenticated.

Connect by signing in (OAuth)

1. Open the provider selector (next to the chat field) or the model Settings panel.
2. Pick the provider and hit connect: your browser opens on the official sign-in screen.
3. Authorize access; back in the app the provider shows as Connected.

Tokens are stored on your machine and refresh themselves when they expire; if a refresh ever fails (because you revoked access, for example), the app will ask you to sign in again.

Gemini no longer connects with your Google account

Google shut down free access through that channel (Cloud Code Assist): the login worked and then every model failed. The living route for Gemini is Gemini API Key, right below. Anyone who already has an old OAuth profile with their own Google Cloud project keeps it intact.

Connect with an API key

1. Get the key from the provider’s website.
2. In the model settings panel, create or edit that provider’s profile and paste the key.
3. Save: the status turns to Connected.

The panel also lets you manage credentials, parameters and limits per provider, and keep several profiles (two accounts on the same provider, for instance).

Gemini API Key

Gemini API Key is the direct developer access to Gemini. You get the key from Google AI Studio and paste it into the provider’s profile.

Since version 3.6 it speaks Google’s native API instead of its OpenAI compatibility layer. For you that means:

• Fewer “the model is overloaded”: on a 503 (“high demand”) or a 429, the app retries on its own three times (at 2, 5 and 10 seconds) instead of throwing the error in your face. If there’s still no room, it tells you plainly: “Google reports the model is overloaded: wait a moment and send again”.
• Real reasoning: the Reasoning switch in the composer controls Gemini 3’s native thinkingLevel (Light / Balanced / Deep). With the switch off, the model thinks the bare minimum instead of ignoring the setting.
• Errors in your language with a hint of what to do, instead of the raw API dump.

Models offered: Gemini 3.8 Flash, 3.1 Pro, 3.6 Flash, 2.5 Pro and 2.5 Flash (AI Studio). If your key gives access to more, Google’s live list wins.

xKiro

xKiro is a gateway: a single key (sk-xt-…) for more than 110 models from GPT, Claude, Gemini, DeepSeek, Qwen, GLM, Kimi, MiniMax, Grok, Mistral, NVIDIA and others — with more than 40 free models on its free tier.

1. Create the key at xkiro.com/dashboard/api/keys.
2. Paste it into the xKiro profile in the model panel.
3. Pick a model. Identifiers carry a vendor prefix (openai/gpt-5.6-sol, anthropic/claude-sonnet-5, minimax/minimax-m3:free): without it, the API answers 404.

Tip

Models marked Free (the ones ending in :free) work on every plan, including the free one: it’s the cheapest way to try several vendors without opening an account with each.

Local models with Ollama or LM Studio

If you already use Ollama or LM Studio, Context Code detects them as just another provider: select one and you’ll see the list of models you have installed (in Ollama the list is dynamic). There is nothing to authenticate. If you’d rather not install anything external, Context Forge plays the same role inside the app.

Switch models on the fly

The provider and model selector lives next to the input field. Changing it does not restart the conversation: the next message goes out with the new model. It’s the natural way to work cheap by default and step up to a powerful model only when the task calls for it.

You can also do it by talking, without touching the mouse: “switch to model X”, “move over to Gemini” — see voice commands.

Tip

If a chat stops responding after you switch providers, check that the active profile for that provider has credentials (status Connected). It’s the number one cause — see Troubleshooting.

More from this screen

The list goes on: local engines (Context Forge) and third-party ones live in the same dropdown.

The right side of the chip switches models only, without leaving the provider.