The speech API that knows who's talking

# 1 — upload
curl -X POST https://rewayai.ai/api/pipeline/upload \
  -H "Authorization: Bearer rw_..." \
  -F "file=@meeting.mp3"
# → {"key": "u42/ab12cd34.mp3", ...}

# 2 — create a job (asr + diarization + attribution)
curl -X POST https://rewayai.ai/api/pipeline/jobs \
  -H "Authorization: Bearer rw_..." \
  -d '{"input_key": "u42/ab12cd34.mp3"}'
# → {"id": 123, "status": "created", ...}

# 3 — poll; stages land incrementally
curl https://rewayai.ai/api/pipeline/jobs/123/stages \
  -H "Authorization: Bearer rw_..."
# → "merged": [{"speaker": "SPEAKER_00",
#      "text": "Welcome everyone..."}],
#    "llm": {"speakers_meta": {"SPEAKER_00":
#      {"name": "Sarah Chen", "confidence": 0.92}}}

import requests

API = "https://rewayai.ai/api/pipeline"
H = {"Authorization": "Bearer rw_..."}

# upload, then create a job
key = requests.post(f"{API}/upload", headers=H,
    files={"file": open("meeting.mp3", "rb")}).json()["key"]
job = requests.post(f"{API}/jobs", headers=H,
    json={"input_key": key}).json()

# poll until done, then read named segments
stages = requests.get(f"{API}/jobs/{job['id']}/stages",
                      headers=H).json()
names = {s: m["name"] for s, m
         in stages["llm"]["speakers_meta"].items()}
for seg in stages["merged"]:
    print(names.get(seg["speaker"]), "—", seg["text"])

import asyncio, json, websockets

async def stream(pcm_chunks):  # 16 kHz mono PCM16
    uri = "wss://rewayai.ai/api/pipeline/stream"
    async with websockets.connect(uri) as ws:
        await ws.send(json.dumps(
            {"token": "rw_...", "language": "en"}))
        async for msg in ws:
            ev = json.loads(msg)
            # {"type": "partial"|"final", "text": ...}
            print(ev["type"], ev.get("text", ""))