Rate Limits

BRAHMA requests are rate limited per organization — the organization that owns the API key — on two dimensions at once, following OpenAI’s rate-limit semantics. Exceeding either limit returns 429 until the corresponding window resets.

The staging base URL is api.stage.peachdeskai.com. The production base URL api.peachdeskai.com applies once production DNS is live.

Limits

Limit	Unit	Default	Applies to
RPM	requests per minute	`60`	Every request to the BRAHMA `v1/*` routes
TPM	tokens per minute	`100000`	Estimated tokens per chat-completion request

Defaults apply per organization and can be raised for your deployment — contact your PeachDesk administrator.

How TPM is counted

The TPM charge for a request is taken up front, before the completion runs:

TPM charge = max(max_tokens, ceil(request characters / 4))

That is, the larger of the max_tokens you request and a characters-divided-by-four estimate of your prompt. Setting max_tokens far higher than you need therefore burns TPM budget even when the completion comes back short — request only the output length you need.

Rate-limit headers

Every response — success or failure — carries your organization’s current standing in the response headers:

Header	Example	Meaning
`x-ratelimit-limit-requests`	`60`	Your RPM limit
`x-ratelimit-remaining-requests`	`59`	Requests remaining in the current minute
`x-ratelimit-reset-requests`	`1s`	Time until the request window resets
`x-ratelimit-limit-tokens`	`100000`	Your TPM limit
`x-ratelimit-remaining-tokens`	`99230`	Tokens remaining in the current minute
`x-ratelimit-reset-tokens`	`6m0s`	Time until the token window resets

Reset values are human-readable durations such as 1s or 6m0s. Read remaining before dispatching bursts to stay under the limit proactively rather than reactively.

The 429 response

When a limit is exceeded, the API returns 429 with a retry-after header (seconds to wait) and an OpenAI-format error body. type is tokens for a TPM breach and requests for an RPM breach:

{
  "error": {
    "message": "Rate limit reached for brahma-30b in organization org-7f3a on tokens per min (TPM): Limit 100000, Used 99230, Requested 4000. Please try again in 6m0s.",
    "type": "tokens",
    "param": null,
    "code": "rate_limit_exceeded"
  }
}

Handling 429s: exponential backoff

Retry 429 responses with exponential backoff and jitter — honor retry-after as the floor, and randomize so concurrent clients do not retry in lockstep.

import random
import time

import requests

URL = "https://api.stage.peachdeskai.com/api/v1/voice-engine/llm/v1/chat/completions"

def complete(payload, max_retries=5):
    delay = 1.0
    for attempt in range(max_retries):
        resp = requests.post(
            URL,
            headers={"X-API-Key": "bibha_your_api_key"},
            json=payload,
        )
        if resp.status_code != 429:
            resp.raise_for_status()
            return resp.json()
        retry_after = float(resp.headers.get("retry-after", delay))
        time.sleep(max(retry_after, delay) + random.uniform(0, 1))
        delay *= 2
    raise RuntimeError("rate limited after retries")

const URL =
  "https://api.stage.peachdeskai.com/api/v1/voice-engine/llm/v1/chat/completions";

async function complete(payload, maxRetries = 5) {
  let delay = 1000;
  for (let attempt = 0; attempt < maxRetries; attempt++) {
    const resp = await fetch(URL, {
      method: "POST",
      headers: {
        "X-API-Key": "bibha_your_api_key",
        "Content-Type": "application/json",
      },
      body: JSON.stringify(payload),
    });
    if (resp.status !== 429) return resp.json();
    const retryAfter =
      Number(resp.headers.get("retry-after") ?? 0) * 1000 || delay;
    await new Promise((r) =>
      setTimeout(r, Math.max(retryAfter, delay) + Math.random() * 1000)
    );
    delay *= 2;
  }
  throw new Error("rate limited after retries");
}

Keep max_tokens close to the output length you actually need — it is charged against TPM up front — and spread batch workloads across the minute instead of front-loading them.

Resources

Authentication & Errors

Voice Engine API

Bibha BRAHMA LLM

Bibha VAANI TTS

Bibha LIPI STT

Limits

How TPM is counted

Rate-limit headers

The 429 response

Handling 429s: exponential backoff

​Limits

​How TPM is counted

​Rate-limit headers

​The 429 response

​Handling 429s: exponential backoff

Limits

How TPM is counted

Rate-limit headers

The 429 response

Handling 429s: exponential backoff