# Setup instructions for every major AI provider – US, Chinese, and local > Source: > Published: 2026-07-30 21:58:34+00:00 Connecting an AI service is what puts models in Big-AGI: where to start, one key for many models, every service in the app, the model each job uses, and models on your own machines. Big-AGI starts empty: connecting an AI service is what fills it with models. One key from one service is enough to start: no minimum spend, and a second service costs nothing to add. A key is not a chat subscription: AI services sell their consumer chat product and their API as two accounts with two balances. A ChatGPT Plus or Claude Pro subscription therefore carries no key, and Anthropic [states it directly](https://support.claude.com/en/articles/9876003-i-subscribe-to-a-paid-claude-ai-plan-why-do-i-have-to-pay-separately-for-api-usage-on-console). The key comes from the service's developer console, on its own balance. Open one of the four the app offers first: [OpenAI](/docs/connect-openai), [Anthropic](/docs/connect-anthropic), [Gemini](/docs/connect-gemini) or [OpenRouter](/docs/connect-openrouter). Want a specific vendor's own models - go to that vendor. Want the fewest steps to a working key - OpenRouter. Want to try without a payment method - [Gemini](/docs/connect-gemini) has a FREE tier, on its own terms about what you send. Those four are the **Featured** group in the add-service list, and the same four the first-run setup offers. Everything else is one click further. Three steps cover it: create the key in the service's console, paste it into **Models** (`Ctrl + Shift + M` ), and that service's models fill the list - the signal it worked ([your first chat](/docs/first-chat)). An aggregator does exactly that, and [OpenRouter](/docs/connect-openrouter) is the one in the list: one account, one balance, a catalogue spanning the major vendors' models. Big-AGI carries a one-click handoff for it - **Link OpenRouter Key** in that panel sends you to OpenRouter to authorise and returns with the key installed. The trade is a middleman: one bill, one catalogue, and a list filtered to curated model families by default. A gift icon in the add-service list marks services that publish some models at no charge. Anything outside the wall that speaks the OpenAI API connects too: [custom endpoints](/docs/connect-custom-endpoints). Support for a new service is written in the open-source codebase first and reaches the hosted site on promotion. Big-AGI assigns a model per job rather than per service, across everything you have connected: | Job | Used for | Default | |---|---|---| Chat | Conversations - changeable per message, before sending | Auto | Code | Applying code changes | Auto | Fast | Titles, summaries, small utility calls | Auto | Vision | Reading images | Auto | **Auto** resolves the moment it is read, and a hand-curated pick wins where one exists. Pin a job with the **Default / Model for** buttons in a model's gear; the dropdown above a chat changes that chat only. [Ollama](/docs/connect-ollama), [LM Studio](/docs/connect-lmstudio) and [LocalAI](/docs/connect-localai) connect like any other service: point Big-AGI at the server's address, and its models join the same list - no key, no account, no per-token bill. With [Direct Connection](/docs/feature-direct-connection) off, the Big-AGI fast edge servers call the address you gave; a cloud server cannot reach an address on your computer. With it on, your browser calls the address directly - which is why the panel recommends it for local services. **Direct Connection works only with your API key stored in the browser**, and with an AI service that permits direct browser calls (CORS). Where it cannot be used, requests route through the Big-AGI fast edge servers instead - everything still works, within the standard upload size and time limits. Running locally gives you: BIG-AGI Resources © 2026 [Token Fabrics](https://www.tokenfabrics.com)·Built with passion in San Diego