Making requests - cloro
Guides
Making requests
Compare the two cloro request modes — synchronous and asynchronous — including latency, concurrency limits, and guidance on which mode to use.
cloro offers two ways to call the API: synchronous requests, where you wait for the result on the same connection, and asynchronous requests, where you submit a task and retrieve the result later.
Navigate to header: Synchronous requests Synchronous requests
Synchronous endpoints return the full result in the same HTTP response, so the request has to fit inside your
client’s timeout
. See
Synchronous requests
for the request and response structure, common parameters, and code examples.
Navigate to header: Asynchronous requests Asynchronous requests
Asynchronous endpoints accept a task, return a
taskId
immediately, and process the work in the background. You retrieve the result via webhook (recommended) or by polling. See
Asynchronous requests
for the two-step submit-and-fetch flow, task states, queue limits, and request prioritization.
Navigate to header: When to use which When to use which
| Use case | Choose |
|---|---|
| Time-sensitive request that needs an immediate answer | Synchronous |
| Serverless environment with short execution time limits | Asynchronous |
| Submitting large batches | Asynchronous |
| Building a resilient system that shouldn’t rely on a long-running connection | Asynchronous |
| Small batches needing immediate results | Synchronous with concurrent workers |
Navigate to header: Common questions Common questions
Navigate to header: How long should one request take? How long should one request take?
Typical end-to-end times, measured over the last seven days of production traffic:
| Surface | Median | 9 in 10 under |
|---|---|---|
| 5s | 19s | |
| Google News | 8s | 45s |
| AI Mode | 15s | 22s |
| ChatGPT | 21s | 38s |
| Gemini | 33s | 41s |
A request to an AI assistant (ChatGPT, Gemini) drives a real browser session (
how cloro retrieves data
): it loads the surface, submits the prompt, waits for the answer to finish streaming and parses the rendered page, often over several page loads when the surface defers its sources or product cards. A Google SERP is one page load, so it lands in seconds. An AI surface has to generate its answer before there is anything to read, and that generation time is the bulk of the wait. The spread moves with the provider’s own load, so treat these as the shape rather than a guarantee.
Elmo
publishes an independent comparison of cloro and other scraping APIs for these engines, re-measured every six hours on the same prompts.
Navigate to header: Why are my requests slower than expected? Why are my requests slower than expected?
Beyond the generation time above, latency comes from:
Queue depth (async requests only): set by your plan’s concurrency limit and your current queue size
Geographic routing: requests route through region-specific infrastructure, which may differ in latency depending on the country parameter you specify
Provider load: peak hours on the provider’s side lengthen generation
If a batch takes far longer than one request times the number of prompts, the cause is almost always concurrency rather than any single request: on one slot the prompts run one after another.
Upgrading your plan
raises concurrency, which drains the async queue faster. To measure a single request, read its
latency header
.
Navigate to header: Can I get faster processing for high-volume workloads? Can I get faster processing for high-volume workloads?
Yes — the larger the plan, the greater the concurrency assigned to it. For async requests, higher concurrency means more tasks processed at once and shorter queue waits. For sync requests, it means you can send more simultaneous requests without hitting limits.
Navigate to header: How do I submit many requests in one call? How do I submit many requests in one call?
Use the
batch endpoint
. It takes
up to 500 tasks
in a single HTTP request and validates each task independently, so one bad task doesn’t block the rest. cloro queues the tasks, within your
queue limit
, and runs them
up to your plan’s concurrency limit
; retrieve results with webhooks or polling.
For which pattern to use once you’re submitting at volume — async with webhooks, or sync with concurrent workers — see
the concurrency guide
, which has code examples for both.
Was this page helpful?
Ask a question...
up to your plan’s concurrency limit
Navigate to header: Synchronous requests
Navigate to header: Asynchronous requests
Navigate to header: When to use which
Navigate to header: Common questions
Navigate to header: How long should one request take?
Navigate to header: Why are my requests slower than expected?
Navigate to header: Can I get faster processing for high-volume workloads?
Navigate to header: How do I submit many requests in one call?