Realtime (WebSocket)

The Realtime endpoint opens a persistent WebSocket connection for low-latency, streaming speech-to-speech and multimodal sessions — audio and text flow in both directions over one socket, rather than one request/response per turn.

GET wss://router.accels.tech/v1/realtime?model=<model>

Parameter	In	Required	Notes
model	query		Realtime-capable model id, e.g. gpt-realtime

Authenticate with your sk- key as a Bearer token on the connection request (Authorization: Bearer sk-...). Once connected, exchange the standard Realtime session events (session config, audio append, response create, etc.).

Example

import websockets, json

url = "wss://router.accels.tech/v1/realtime?model=gpt-realtime"
headers = {"Authorization": "Bearer sk-..."}

async with websockets.connect(url, additional_headers=headers) as ws:
    await ws.send(json.dumps({
        "type": "response.create",
        "response": {"instructions": "Say hello in one short sentence."},
    }))
    async for message in ws:
        print(message)

Notes

• Uses the standard Realtime event protocol; existing Realtime SDKs work by pointing the base URL at router.accels.tech.
• Realtime-capable models are listed in the model marketplace.