Ramsa is an API for turning text into authentic Emirati speech. Core concepts: agents (characters with voices and accents), voices, sessions, and streaming.

Each agent has an agent_id and a voice_id that you use in your requests.

Install the SDK, then synthesize your first Emirati line in under five minutes.

npm install @ramsa/sdk

import { Ramsa } from "@ramsa/sdk";
const ramsa = new Ramsa({ apiKey: process.env.RAMSA_KEY });

const audio = await ramsa.speak({
  agent: "salem",
  text: "هلا والله، نوّرت رمسة",
});

Pass your API key in the request header. Create and rotate keys from the dashboard, and keep test and live environments separate.

curl https://api.ramsa.ai/v1/speak \
  -H "Authorization: Bearer rmsa_live_..." \
  -H "Content-Type: application/json" \
  -d '{ "agent": "salem", "text": "..." }'

The /speak endpoint takes an agent, text, and optional voice settings (speed, pitch, emotion, stability, dialect intensity).

await ramsa.speak({
  agent: "moza",
  text: "هلا فيك! تبي مساعدة؟",
  settings: { speed: 60, emotion: 70, dialectIntensity: 75 },
  format: "mp3",
});

For real-time interactions, consume the event stream: interim ASR results, then tokens, then audio chunks. Wire it to a push-to-talk button.

for await (const chunk of ramsa.stream({ agent: "salem", input })) {
  if (chunk.type === "token") render(chunk.text);
  if (chunk.type === "audio") play(chunk.audio);
}

The API returns standard HTTP status codes. 401 invalid auth, 429 rate limited, 400 bad request. Retry with exponential backoff on 429/5xx.