Skip to main content

Rate limits

Four tiers, applied by endpoint group.

EndpointsLimit
Authentication — /auth/login, /auth/register, /auth/refresh10 requests / 15 minutes
Widget chat — /chat/:id/message20 requests / minute
Ingestion and uploads — adding sources, refreshing, recrawling, image uploads15 requests / minute
Everything else under /api120 requests / minute

Exceeding a limit returns 429 Too Many Requests.

Why authentication is so tight​

Ten attempts per fifteen minutes is deliberately restrictive — it is the defence against password guessing.

Do not build a retry loop against /auth/login. If a login fails, the credentials are wrong; retrying will lock you out of the endpoint without ever succeeding.

Log in once and reuse the session cookies for the rest of your run.

Request size caps​

LimitValue
JSON and form request bodies1 MB
Document upload25 MB
Image upload2 MB

The 1 MB body cap applies to JSON. File uploads use multipart/form-data and are governed by their own limits, not this one.

A very long pasted-text source can exceed 1 MB. Split it into several sources — which is better practice anyway, because focused sources retrieve more accurately than one enormous one.

Handling a 429​

Back off and retry. Do not retry immediately in a tight loop.

async function withRetry(fn, attempts = 4) {
for (let i = 0; i < attempts; i++) {
const res = await fn();
if (res.status !== 429) return res;
// 1s, 2s, 4s, 8s
await new Promise(r => setTimeout(r, 1000 * 2 ** i));
}
throw new Error('Rate limited after retries');
}

Staying under the limits​

Bulk-loading content. Ingestion allows 15 requests per minute. Adding fifty sources means pacing them — roughly one every four seconds, or batches with a pause between.

Polling for training status. GET /chatbots/:id/status falls under the general 120/minute limit, so once every few seconds is comfortable. Crawls take minutes; polling more often than every five seconds tells you nothing new.

Fetching leads or history on a schedule. Once every few minutes is ample. A tight loop wastes your allowance on unchanged data.

Message allowance is not a rate limit​

Separately from these per-minute limits, your account has a monthly message allowance from your plan.

When it runs out, widget messages fail with 429 Message limit reached. Unlike a rate limit, waiting does not clear this — it resets with your billing period, or when you upgrade.

The two are distinguishable by the error message:

MessageMeaningFix
Message limit reachedMonthly allowance exhaustedUpgrade, or wait for the reset
(rate limiter response)Too many requests too quicklyBack off and retry

Chatbot limit​

Creating a chatbot beyond your plan's allowance returns 403 Chatbot limit reached for your plan. That is a 403, not a 429 — it is a permission outcome, not a throughput one, and retrying never helps.