API Limits

The Speechify API enforces three kinds of limit:

Rate and concurrency limits are enforced per account, not per API key. On Build audio they are shared across every synthesis endpoint - /v1/audio/speech, /v1/audio/stream, and /v1/audio/stream/with-timestamps all draw from one budget, so a request to any of them counts against the same rate and concurrency ceiling. Exceeding either returns 429 Too Many Requests.

Character limits

Rate limits

Rate limits differ by product because the workloads differ. Build audio is cost-per-call. Agents is chatty interactive traffic.

Build audio

Applies to every Build audio synthesis endpoint - /v1/audio/speech, /v1/audio/stream, and /v1/audio/stream/with-timestamps - which share one per-account rate budget.

Agents

Applies to every Agents endpoint under /v1/agents/*: agents, conversations, knowledge bases, tools, tests, memories, audio assets, batch calls, IVR menus, and telephony.

Concurrency limits

Concurrency limits cap the number of simultaneous in-flight requests per account.

Build audio

Applies to every Build audio synthesis endpoint - /v1/audio/speech, /v1/audio/stream, and /v1/audio/stream/with-timestamps - which share one per-account concurrency budget.

Agents

Applies to the authenticated Agents endpoints listed above. The primary target is POST /v1/agents/{id}/conversations, which allocates a live-call session.

Reading your budget

Every response on a rate-limited endpoint carries the request-rate budget headers, so clients can pace themselves before hitting 429:

The same values are mirrored as X-RateLimit-* for clients predating the IETF draft names.

Every Speechify-specific header follows the same rule: the canonical name is un-prefixed and namespaced (Speechify-Request-Id, Speechify-Version, Speechify-Signature), and any pre-2026 X- spelling is still accepted and still emitted alongside it until 2027-07-24.

Handling 429 responses

When you exceed rate or concurrency limits, the API returns 429 Too Many Requests with a Retry-After header. The error body names the limit your plan allows and links back to this page (error.docs_url); the two cases are distinguishable by error.code: rate_limited (requests per second) vs concurrency_limit_reached (too many at once).

Need more headroom? Every limit above rises with your plan - upgrade in the console under Billing, or contact us for Enterprise terms.

import time
from speechify import Speechify
client = Speechify()
def generate_with_retry(text, max_retries=3):
    for attempt in range(max_retries):
        try:
            return client.audio.speech(
                input=text,
                voice_id="geffen_32",
                model="simba-3.2",
                audio_format="mp3",
            )
        except Exception as e:
            if "429" in str(e) and attempt < max_retries - 1:
                time.sleep(2 ** attempt)
            else:
                raise

Processing long texts

For texts exceeding 20,000 characters, split into chunks and process sequentially:

def split_text(text, max_chars=19000):
    """Split text at sentence boundaries within the character limit."""
    chunks = []
    current = ""
    for sentence in text.split(". "):
        if len(current) + len(sentence) + 2 > max_chars:
            chunks.append(current.strip())
            current = sentence + ". "
        else:
            current += sentence + ". "
    if current.strip():
        chunks.append(current.strip())
    return chunks

FAQ