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

# Comms Router

> How Comms meters workspace models for hosted agents and the chat completions API.

Comms Router is the workspace model layer. Your app (or your hosted agent) spends tokens. Those tokens are billed per model — no plan includes AI usage.

You do not paste a third-party model key into Comms for the workspace path. Bring-your-own providers remain available on an agent if you add one.

## What is metered

| Source                                | When                          |
| ------------------------------------- | ----------------------------- |
| Hosted agent replies                  | After a successful run        |
| `POST /api/v1/comms/chat/completions` | After a successful completion |

Both write to the same `ai_token` meter for the billing period.

## Pricing

Each model has its own input and output rate (\$ / 1M tokens).

* Catalog rate is charged as-is.
* When a model is discounted versus its published list price, the customer rate is **5% below** the catalog rate.
* List models and rates with `GET /api/v1/comms/ai/models`.

Messaging Services Agreement workspaces invoice Comms Router line items in arrears with dedicated-line fees and message overage.

## API

1. Mint a key with the **AI** scope (`comms_ai`) on [API keys](https://comms.osis.co/dashboard/api).
2. Call [chat completions](/messages-api/chat-completions).
3. Read [AI usage](/messages-api/ai-usage) for this period's token meter.

The dashboard **Developers → AI** tab has the same endpoint and a live meter.

## Compatibility

The completions endpoint is OpenAI-shaped:

* `messages[]` with `role` + `content`
* `model`, `temperature`, `max_tokens`
* response `choices[0].message.content` and `usage`

Point an OpenAI SDK `baseURL` at `https://osis.co/api/v1/comms` and set `apiKey` to your Comms key. Streaming is not supported yet (`stream: true` returns 400).
