> For the complete documentation index, see [llms.txt](https://codemash.gitbook.io/codemash/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://codemash.gitbook.io/codemash/ai/products/llm-providers.md).

# LLM providers

When and why to connect an LLM provider to your project, which ones run the chat today, and how Norbix picks the model.

Every AI feature of your app needs a model to think with. An **LLM** (large language model) is that model — GPT, Llama, Mistral and others — and a provider is the company or server that runs it. You want to choose the provider yourself (price, speed, data region), use your own key, and change it later without touching your app.

## How Norbix models it

An **LLM integration** stores one provider for one environment of the project: the provider, a name, the API key (kept encrypted, never shown again), an optional **Endpoint** and a **Default Model**. You can add several — a cheap fast model and a strong one — and each assistant can use its own.

Which model a chat turn runs on, first match wins:

| Order | Where it is set                                                                              |
| ----- | -------------------------------------------------------------------------------------------- |
| 1     | The assistant's own **LLM integration** and **Model** (Project → Settings → AI → Assistants) |
| 2     | The project's **Default LLM** (Project → Settings → AI → Default LLM)                        |
| 3     | The LLM integration marked **DEFAULT** in the list (the first one you add gets it)           |

| Provider                                                         | Works in the chat today | Notes                                                   |
| ---------------------------------------------------------------- | ----------------------- | ------------------------------------------------------- |
| [OpenAI](/codemash/cloud/ai/llm/providers/openai.md)             | yes                     | Also any OpenAI-compatible server through **Endpoint**. |
| [Groq](/codemash/cloud/ai/llm/providers/groq.md)                 | yes                     | Fast open models.                                       |
| [Mistral](/codemash/cloud/ai/llm/providers/mistral.md)           | yes                     |                                                         |
| [OpenRouter](/codemash/cloud/ai/llm/providers/openrouter.md)     | yes                     | Many models behind one key.                             |
| [Grok](/codemash/cloud/ai/llm/providers/grok.md)                 | yes                     |                                                         |
| [Anthropic](/codemash/cloud/ai/llm/providers/anthropic.md)       | no                      | Saves and tests, but every chat turn fails.             |
| [Google](/codemash/cloud/ai/llm/providers/google.md)             | no                      | Same.                                                   |
| [Ollama](/codemash/cloud/ai/llm/providers/ollama.md)             | no                      | Same.                                                   |
| [Norbix model](/codemash/cloud/ai/llm/providers/norbix-model.md) | no                      | The managed-service card; saving is refused today.      |

{% hint style="warning" %}
Anthropic, Google and Ollama integrations save and pass **Test integration**, but the chat has no connector for them yet, so every turn fails with *Error while running chat turn*. Use OpenAI, Groq, Mistral, OpenRouter or Grok — or reach Claude and Gemini models through OpenRouter.
{% endhint %}

**Test integration** only checks that the key is stored; it does not call the provider. A wrong key shows up on the first chat turn. Many providers accept at most 128 tools in one request, so give an assistant only the toolsets it needs.

When to add more than one:

* **Per environment** — TEST and PROD keep separate integrations, so a test key never spends production money.
* **Per assistant** — a small model for a support assistant, a large one for a report writer.
* **Fallback by hand** — if a provider has an outage, switch the default on Settings → AI; apps and assistants keep working without a release.

## Where you use it

* Cloud: [LLM integrations](/codemash/cloud/ai/llm.md) · [Connect a provider](/codemash/cloud/ai/llm/providers.md) · [Test](/codemash/cloud/ai/llm/test.md) · [Set the default LLM](/codemash/cloud/ai/llm/set-default.md) · [Disable or delete](/codemash/cloud/ai/llm/disable.md)
* Project → Settings → [AI](/codemash/cloud/project/settings/ai.md) — default LLM, assistants, plans
* [AI chat](/codemash/ai/products/ai-chat.md) — where the model is used
* SDK: [AI](/codemash/sdks/ai.md) · API: [AI](/codemash/api-reference/ai.md)