> ## Documentation Index
> Fetch the complete documentation index at: https://docs.octomerge.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Open LLM Router apps

> Connect one of your Open LLM Router apps in Settings → Models, choose its models as your default, refresh it when the app changes, and revoke r5d's access.

[Open LLM Router](https://open-llm-router.r5d.app) keeps your model providers' API keys and subscriptions, and groups them into **apps**: each app offers models with their efforts and decides which of your accounts serves them. You can connect an app to r5d as one of your AI connections. Its models then appear in each chat's model menu and to agents, like those of any connection.

Requests run through the app and use the quota or cost of its accounts. Your keys and subscriptions stay in Open LLM Router: r5d gets access to that one app, its models and requests through it, and never sees your other apps, accounts, keys or logs.

> [!NOTE]
>
> Open LLM Router apps are offered only where your r5d installation is set up for Open LLM Router. Without it, **Settings → Models** offers no router apps. Apps connected before stay listed as not configured: their requests fail, and you can rename or delete them.

## Connect an app

1. Open **Settings → Models** and choose **Add connection → Open LLM Router app**, then **Connect**.
2. Open LLM Router opens. Sign in if it asks (with r5d, you are usually signed in already), choose the app and approve r5d's access to it. **Deny** connects nothing.
3. Back in **Settings → Models**, a preview shows what the app offers. Nothing is saved yet: choose **Finish** to add the connection, or **Cancel** to revoke the access you just approved.

Approve within 10 minutes of choosing **Connect**, and finish the preview within 30 minutes; after that, connect again. A preview you closed without finishing waits on the page with a **Review** button until it expires.

The setup checklist's **Model providers** step connects an app the same way: **Add providers** opens Open LLM Router in a window, or a tab on phones, and the checklist stays where it is. When you approve, the window closes and the checklist shows the preview. If your browser blocks the window, open Open LLM Router in a new tab from the link the step offers; you then finish in that tab.

## What the preview shows

* The app's name and description, and **connected as** the Open LLM Router account that approved it.
* **Models** that will be imported, each by its ID with the efforts it accepts and its default effort, its context window and output limit. A model that pins an effort, such as `mine-max`, runs at that effort and takes none of its own. Efforts the app's active account can't run now are listed as not available.
* **Not imported**: models r5d leaves out, with the reason: the app can't run them now, their limits aren't known, or they aren't served over Chat Completions. Fix them in Open LLM Router and refresh later.

When none of the app's models can be imported, the preview says so and offers only **Cancel**, with **Manage in router** to open the app in Open LLM Router.

r5d imports exactly what the app offers: it doesn't invent models or efforts, and a model's efforts and default come from the app. To change them, change the app in Open LLM Router and refresh.

## Your default model

The preview offers **Use *model* as my default**, for the app's own model at its default effort (or its first model). It is ticked when you have no default model yet, and unticked when you have one, so your default changes only when you tick it. Projects and branches keep their own defaults. Connecting an app again never changes defaults.

## Model IDs are unique

Each model ID belongs to one of your connections, so a chat, a default or an agent always names one model.

* Another Open LLM Router app can't import a model ID one of your connected apps already imports: its preview or refresh is refused until you rename the model or the app in Open LLM Router (**Manage in router**). An app's model IDs come from Open LLM Router and can't be renamed in r5d.
* When an app brings an ID that one of your own keys' connections uses, the preview names that connection and offers to rename the key's model, for example to `claude-sonnet-4-6-2`. The renamed model keeps sending its old ID to its provider, so it works as before; your defaults that chose it follow the new ID. **Finish** waits until no ID clashes. Nothing is renamed or dropped without your confirmation.
* A model ID you add to a connection yourself can't be one another connection uses either. The editor flags it as you type and offers a free ID; the model can still be sent to its provider under its own ID (**Sent to the provider as**).

IDs that two connections already shared before this rule are flagged in the list, naming both connections. They keep working; saving either connection's models asks you to rename one.

## Manage a connected app

An app's connection appears in **Settings → Models** with an **Open LLM Router app** badge and its status: **Connected**, **Reconnect needed**, **Disconnecting** or **Disconnected**. Enable it, make it the default and choose its default model as for any connection. Its **⋯** menu opens **Open LLM Router app…**, which shows the app, the account it was approved by, when r5d last read it and its models, with these actions:

* **Refresh** reads the app again and lists what changed: models added, removed or renamed, their efforts and limits, and the app's name. **Apply changes** updates the connection; **Keep current models** leaves it as it is. When nothing changed, it says the app is up to date and records when r5d last read it. Models keep their place across renames. A model a chat or default chose that the app no longer offers is refused when it runs, never swapped for another. Changes behind an unchanged model ID, such as which of your accounts serves it, are Open LLM Router's and need no refresh.
* **Manage in router** opens the app in Open LLM Router, where its models, efforts and accounts are.
* **Disconnect**, **Reconnect**, **Rename** and **Delete**: see below.

The connection's base URL and headers belong to Open LLM Router, so it has no **Edit** form.

## Revoke access

* **Disconnect** revokes r5d's access in Open LLM Router. Requests already running finish; new ones fail with **Reconnect Open LLM Router**, and nothing falls back to another connection. The connection stays, so chats keep their model choice: **Reconnect** and approve the same app to use it again. While Open LLM Router can't be reached, the connection shows **Disconnecting** and isn't used; r5d revokes it as soon as it can.
* **Delete** revokes the access too, and removes the connection.
* In Open LLM Router, the **Account** page's **Connected apps** lists r5d's access to each app; revoking it there shows the connection as **Reconnect needed** in r5d.

## Sign in and link accounts

Open LLM Router signs you in with your r5d account. If you already used Open LLM Router with Google, link the two: sign in to Open LLM Router with Google, open its **Account** page and choose **Link r5d account** under **Sign-in methods**. Then your apps are there when you connect from r5d, and r5d opens Open LLM Router signed in.

## Delete your Open LLM Router account

* **In Open LLM Router:** its **Account** page ends with **Delete account**, which deletes your router account and everything in it: apps, provider accounts with their keys and sign-ins, request logs, usage history, API keys and connected apps. Connections to it in r5d then show **Reconnect needed**.
* **With your r5d account:** on r5d.dev, **Settings → Delete account** deletes your Open LLM Router account too when you sign in to it with r5d. One that also signs in with Google stays, without its r5d sign-in. See [Deleting your account](https://docs.octomerge.dev/get-started/sign-in.md#deleting-your-account).

## When a request fails

A chat's agent stops with a notice when Open LLM Router refuses a request, and says what to do:

* **The app's account or key was refused by its provider**: fix the app's account or key in Open LLM Router (the connection's **Manage in router**), or choose another model, then **Resume**.
* **The app's tunnel is offline**: the app serves a model through a tunnel on one of your computers; start it there, or choose another model.
* **The app's endpoint is on its tunnel computer's network**, and the tunnel wasn't started to allow it: restart the tunnel with the command its **Tunnel** page in Open LLM Router shows.
* **The app's endpoint can't be used as it is**: Open LLM Router doesn't connect to its address directly, it answered with a redirect, or it didn't answer like a model API (often a wrong base URL). Fix the endpoint in Open LLM Router (**Manage in router**), or choose another model.
* **A model the app maps to left its account's catalogue**: update the app in Open LLM Router, or choose another model.
* **Reconnect Open LLM Router**: the connection was disconnected here or ended in Open LLM Router; reconnect it from its menu, or choose another model.

**Test model** in an app's connection and **Refresh** name the same causes.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.