Providers and models

The agent needs a model to think with. A provider is where that model runs: a cloud service such as OpenAI or xAI, AgenC's own hosted models, or a server on your computer such as Ollama. You can connect as many as you like, pick a default for new sessions, and switch any session to another model at any time.

There are four ways to connect:

• An account you already have. Sign in with X to use Grok, or sign in with ChatGPT to use OpenAI models, on your subscription.
• An API key from the provider.
• AgenC hosted models, through your AgenC account and its credits.
• A model on your computer, through Ollama, LM Studio or any OpenAI-compatible server. Nothing needs an account or a key.

All of this is in Settings > Providers.

Connect a provider

Settings > Providers lists every provider, split into Cloud and Local, with a search box ("Search providers: OpenAI, GLM, Ollama…"). Each provider is a card. Its row shows a status dot, a one-line description, and "default" or "connected" when that applies. A card that is not connected offers a quick action: Add key, Choose connection, Detect or Set up. Click the row to open the card.

With an API key

1. Open the provider's card and click Add key.
2. Paste the key in the dialog ("Stored securely on this device.") and click Save.

The key works for new sessions at once. Update key replaces it and Remove forgets it. After you remove a key, the page shows "Changes saved" with Restart daemon: the old key stays in Core's memory until Core restarts. The button works once no task is running ("Finish active tasks before restarting.").

Sign in with X for Grok

The xAI Grok card offers two connection methods: Account · OAuth ("Your X subscription") and API key ("Provider API billing").

1. Choose Account · OAuth, then click Sign in.
2. Your browser opens to sign in through xAI. The card follows along: "Signing in with Grok…", "Finish signing in through xAI.", then "Browser sign-in received. Finishing…".
3. If the browser cannot call back, the card says "The browser callback was unavailable. Continue with a device code." and shows a code to enter on xAI's site.
4. The card then says "Signed in as" with your account, and Sign out.

Grok's Fast speed (xAI priority processing, at twice the standard price) works only with an API key. A Grok account signed in with X does not offer Fast.

Sign in with ChatGPT

The OpenAI card works the same way: Account · OAuth ("Your ChatGPT account") signs in with ChatGPT in your browser, and API key uses OpenAI API billing. A ChatGPT sign-in reaches a smaller set of models than an API key, and the model list shows only those.

Account or key: which one is used

When a provider has both, the card's Connection method heading shows which one Core uses, for example "Using account" or "Using API key (automatic)". To switch, select the other method and click Use account or Use API key. The change applies to new sessions. Existing sessions keep their connection, and both saved credentials are kept.

AgenC hosted models

The AgenC card, at the top, connects your AgenC account: "Connect your AgenC account for hosted models."

1. Click Sign in with AgenC. Your browser opens to sign in.
2. Under Hosted models, the card says whether models are "Available through your AgenC account."
3. Click Use AgenC to use them.

Your credits appear in Settings > Profile, with View history. Requests through AgenC go only to endpoints that keep no data and never train on it.

A model on your computer

The Local group lists Ollama, LM Studio and Custom endpoint (any OpenAI-compatible server).

1. Open the card. The server address is filled in with the usual one on this computer: port 11434 for Ollama, port 1234 for LM Studio.
2. Click Detect. When the server answers, the card says "Server connected" and lists its models.
3. Click Save server if you changed the address. The app saves it and reloads Core.

For a custom endpoint, save its URL and, if it needs one, a key.

The app can also pick, download and start a local model for you. On the welcome screen after setup, choose Local model; see First run and startup. If your default provider is Ollama on this computer and it is installed but not running, the app starts it quietly at launch.

The providers

Cloud: OpenAI, Anthropic, xAI Grok, Google Gemini, Meta, QwenCloud, QwenCloud Token Plan, Cerebras, Ollama Cloud, OpenRouter, DeepSeek, Groq, Mistral, GitHub Models, MiniMax, NVIDIA NIM, Amazon Bedrock, Z.AI, Z.AI Coding Plan, Kimi (Moonshot) and Venice AI.

Hosted: AgenC. Local: Ollama, LM Studio and Custom endpoint.

Venice AI is a preset of the custom endpoint: Use this endpoint fills in its address, and it uses the custom endpoint's key. One custom endpoint is active at a time.

A provider that Core reports but the app does not know yet still appears, with "Reported by the daemon."

Defaults for new sessions

The Defaults group at the top of Settings > Providers sets what new sessions use:

• Default provider: "Used by new sessions; existing sessions keep their model."
• Default model: "From that provider's family."

Each opens the model picker. A connected provider's card also has Choose a model, which makes that provider and model the default. A provider without working credentials cannot become the default; the page says "Default not changed" and why.

To change the model of one session instead, use the pill in its composer; see Writing messages.

What each card shows

Once connected, a card also shows:

• Models: how many models the provider offers, and Choose a model.
• Sub-agents: the effort and speed of sub-agents on this provider; see Sub-agents.
• Zero data retention: what the provider keeps. It reads "always on", "set on the provider's console" or "not offered", with the provider's own note on hover and Details for its page. OpenRouter has a switch instead: with it on, requests go only to endpoints with a zero data retention policy, and a model without one is refused. The switch turns on when you add an OpenRouter key.
• Funds and local spending, for providers used with an API key: the provider balance or key allowance where the provider reports it, with Refresh balance and Open billing; this month's Local spend, counting only what this app recorded; and an optional Monthly warning in US dollars. The warning is advisory: when it is reached, tasks continue.

If an environment variable points a provider at another server, the card explains that the saved key is not sent there and what to change.

Other providers for sub-agents

At the bottom of Settings > Providers, Agents on other providers lets sub-agents run on other connected providers. See Sub-agents.

Where keys are kept

The app encrypts each key with your operating system's secure storage before saving it in its data folder. On macOS, if that storage is not available, the key goes into your login Keychain instead. On Windows and Linux, if secure storage is not available, the app refuses to save the key: "Secure OS key storage is unavailable. The key cannot be stored safely." Account sign-ins are kept by Core. See Data and privacy.