> For AI agents: the complete documentation index is available at /en/llms.txt, the full documentation bundle is available at /en/llms-full.txt.

# How to Configure LLM Providers and Models in Octop

Octop ships built-in presets for a wide range of LLM backends and lets you add any custom endpoint. Providers are managed by admins under **Management → Models**; the models they expose then feed a shared pool that every agent can draw from.

## Add a provider

```text
Management → Models
```


![Model management](/images/en/ui-models.png "Model management")

Pick a built-in preset from **Cloud** or **Local**, enter your API key, and enable it. The provider's models immediately become available to experts under your account. New preset models are synced when Octop starts. Voice, search, and **decision model** settings (used when the product picks among skills, tools, or team routing) live on this same **Models** page.

For an OpenAI-compatible provider, **OpenAI API Mode** can be **Responses (/v1/responses)**. The optional fallback to Chat Completions stays off unless the provider reports that Responses is unsupported.

Search providers on that page include DeepSeek and **Yuanbao**. DeepSeek returns ranked results, summaries, and source links. Yuanbao covers the open web and Tencent content.

If the active model is unhealthy, Octop can fail over to another healthy model from the pool so chats keep moving.

## Built-in provider presets

The **Cloud** tab covers hosted services:

| Preset                                                                                                                      | Protocol             |
| --------------------------------------------------------------------------------------------------------------------------- | -------------------- |
| Tencent Cloud (including Token Plan enterprise and Hy), Kimi, MiniMax, OpenCode, SiliconFlow, Aliyun, Volcano Engine, Zhipu | `openai`, multi-site |
| OpenAI, OpenAI (ChatGPT), OpenRouter, DeepSeek, Google Gemini, Groq, ModelScope, Xiaomi MiMo                                | `openai`             |
| Anthropic                                                                                                                   | `anthropic`          |

Tencent Cloud includes Token Plan, Token Plan Enterprise (CN), Coding Plan, and Hy Token Plan variants. Hy models such as `hy3` expose reasoning status automatically when the provider supports it.

The **Local** tab ships an **Ollama (Local)** preset pointing at `http://localhost:11434/v1`. If you need something else entirely, use **Custom provider** to supply your own base URL and key.


> **NOTE**
> 
> Presets are labelled by the API protocol they speak — `openai` for the OpenAI chat completions spec and `anthropic` for the Anthropic Messages API. Any endpoint that follows one of those specs works as a custom provider.
> 


## The available model pool and auto-routing

Every model from an enabled provider joins the **available model pool** at the top of the page. Octop can route each message automatically, choosing the model that best fits that message. Models you mark with a star are preferred when nothing about the request calls for something else.

Each model in the pool shows its modality and context window, so you can see at a glance which ones handle long inputs.

Use **Test connection** after entering credentials. Octop maps common authentication, quota, rate-limit, and provider-availability failures to localized guidance so you can fix the provider without reading raw upstream errors first.

## Set a model for an agent

Agents fall back to the global default unless you assign a model to them specifically. Override it per agent when you want, for example, a cheaper model for everyday chat and a stronger one for complex work.

Each account can also store a **preferred model** and per-model **reasoning** defaults under personal settings. Those preferences apply to new turns for that user and do not change the shared provider catalog.


> **TIP**
> 
> To configure providers and models from the terminal, use `octop models config`, `octop models list`, and `octop models active` — see [Agent commands](/en/cli/agent-commands.md#octop-models).
> 


## Use Ollama for local models

Enable the **Ollama (Local)** preset under the **Local** tab. No API key is required; Octop talks to your Ollama instance at `http://localhost:11434/v1`.


> **NOTE**
> 
> Ollama must be running before you send the first message to an agent that uses it. Use `octop models ollama-list` and `octop models ollama-pull` to inspect and download local models.
> 


## Knowledge-base models

Knowledge bases can index documents on this machine or through an online embedding model. Administrators pick the model under **Knowledge Bases → Knowledge base settings**, or download a local model on **Models**. See [Knowledge bases](/en/guide/concepts/knowledge.md).

## Volcengine image and video

When a Volcengine (Ark) provider is enabled, you can configure Seedream image and Seedance video generation, test the models, and preview results from the web UI. These generation models are separate from chat completions.
