Rate Limits - Rhizome AI

Current Limits

Limit	Value
Searches per minute, per account (v3)	30
Document reads per minute, per account (v3)	100
Other requests per minute, per API key	50
Searches running at once, per account (v3)	1
Documents per search (`limit`, v3)	50
Pages per document read (v3)	100
Max results per request (`top_k`, legacy endpoints)	100

An account’s search and document-read limits cover all of its API keys. A per-minute limit counts the requests of the last 60 seconds. A request rejected with 429, or a search answered 503 because the server is busy, doesn’t count, so retrying doesn’t keep you limited.

Rate Limit Headers

Every response to a request with a valid key carries:

Header	Value
`X-RateLimit-Limit`	Requests allowed per minute for this endpoint: `30` for search, `100` for document reads, `50` for the rest.
`X-RateLimit-Remaining`	Requests left in the current minute, this one counted.
`X-RateLimit-Reset`	Seconds until the oldest request in the minute stops counting and frees a request.

When you go over a per-minute limit, the API answers 429 Too Many Requests with a Retry-After header, in seconds. On API v3 the body is:

{
  "error": {
    "code": "rate_limited",
    "message": "Rate limit exceeded. Maximum 30 searches per minute.",
    "request_id": "req_3f9c2a7d1e8b4c6a0f5d2e1b"
  }
}

The legacy endpoints send {"error": "Rate limit exceeded. Maximum 50 requests per minute."} with the same headers.

One search at a time

A v3 search runs one at a time per account, across all of the account’s API keys. A second search sent while the first is still running waits up to 2 seconds for it, then answers 429 with the code concurrent_search and a Retry-After header. Send searches one after another, not in parallel. Reading documents and listing datasets have no such limit. When too many searches are running on the server, a search answers 503 Service Unavailable with the code busy and a Retry-After header. A 429 means your client should slow down; a 503 means the server is busy. Either way, wait Retry-After seconds and retry. See Errors for every code.

Best Practices

Wait as long as Retry-After says

When you receive a 429 or a 503, wait Retry-After seconds before retrying. Without the header, double the wait with each retry. To stay under the limit in the first place, pace requests by X-RateLimit-Remaining and X-RateLimit-Reset.

import time
import requests

def make_request_with_retry(url, headers, json, max_retries=5):
    for attempt in range(max_retries):
        response = requests.post(url, headers=headers, json=json)

        if response.status_code in (429, 503):
            # Retry-After says how long to wait; otherwise 1, 2, 4, 8, 16 seconds
            wait_time = int(response.headers.get("Retry-After", 2 ** attempt))
            print(f"Waiting {wait_time}s...")
            time.sleep(wait_time)
            continue

        return response

    raise Exception("Max retries exceeded")

Batch your requests

Instead of making many small requests, ask for more results per request: limit on v3 search, top_k on the legacy endpoints. Every v3 search request reruns the whole search, so each page costs the same.

# ❌ Bad: 50 requests for 50 documents
for i in range(50):
    search(keywords, limit=1, offset=i)

# ✅ Good: 1 request for 50 documents
search(keywords, limit=50, offset=0)

Cache responses

Cache API responses to avoid making duplicate requests for the same data.

Need Higher Limits?

Contact support@rhizomeai.com to discuss custom rate limits for your use case.

​Current Limits

​Rate Limit Headers

​One search at a time

​Best Practices

​Need Higher Limits?

Current Limits

Rate Limit Headers

One search at a time

Best Practices

Need Higher Limits?