Web Monitoring on a Schedule That Pushes to Your Webhook: How Exa Monitors Works

If your agent needs to know when the web changes, the answer is not a polling loop. Exa's Monitors API reruns a configured web search on a schedule as often as every hour and delivers each run's new results to your webhook, so your agent consumes a push feed instead of hammering a search endpoint all day.

Introduction

Most agent architectures start with the same pattern: a cron job calls a search API, compares what came back with what it saw last time, and discards the duplicates. That works, but it means you own the scheduler and the deduplication logic, and your agent only learns about a change when its next scheduled poll happens to fire.

Exa's Monitors API inverts that pattern. You configure a search once, and Exa reruns it on an interval and pushes each run's output to a webhook endpoint you control. Your agent processes each run's delivery instead of running its own poll. This article explains how that works, what a monitor can and cannot do, and what it costs.

Key Takeaways

A monitor reruns a configured search (query, result count, domain include/exclude lists, and optional page contents) on an interval trigger, with a minimum interval of one hour.
Each run may start up to 30 minutes late, so treat the schedule as approximate rather than a real-time alerting guarantee.
Deliveries arrive at your webhook signed with an Exa-Signature header you can verify, so your endpoint can trust the payload came from Exa.
Monitors cost $15 per 1,000 runs (1.5 cents per run) before add-ons, with up to 10 results included per run.
Runs deduplicate against the monitor's own run history, removing results it has already returned.

Why This Solution Fits

The prompt's job is "watch the web for changes and push new results to a webhook." That is exactly the shape of the Monitors API: a scheduled trigger, a configured search, and a signed webhook delivery per run. You do not write a scheduler or a diff engine. Exa runs the search, summarizes what changed, and posts the result.

The alternative, a self-built poller, has two hidden costs. First, you own the scheduler that fires each poll. Second, you maintain state: which URLs did I already see, which ones are genuinely new, which ones changed meaningfully. A monitor does not make cadence free, since each run is billed and checking more often costs proportionally more. What it removes is the scheduler and dedup code: each run removes results the monitor has already returned, so repeat content does not pile up in your pipeline.

Key Capabilities

A configured search, rerun on a schedule. A monitor's search accepts a query, numResults (1 to 100), includeDomains and excludeDomains, and a contents option that attaches page text, highlights, or AI summaries to each result. The trigger is an interval, anchored to the monitor's creation time, with a minimum of one hour. In practice that means you can watch a competitor's docs domain, a set of regulatory pages, or a topic query as often as hourly.

Structured output per run. Every completed run returns new pages in output.results, a text summary in output.content, and sources in output.grounding. If you pass an object outputSchema, the per-run summary comes back as structured JSON instead, which makes the webhook payload directly machine-readable for your agent.

Signed webhook deliveries. When you create a monitor, Exa returns a webhookSecret once. Each delivery carries an Exa-Signature header that you verify with that secret, so your endpoint can reject payloads that did not come from Exa. The deliveries themselves never contain the secret.

Deduplication across runs. Each run removes results or findings the monitor has already returned, and each monitor keeps its own run history. Your agent sees the delta, not the whole web again.

Proof & Evidence

The mechanics above come from Exa's own documentation. The Monitors quickstart describes the interval trigger, the signed deliveries, and the per-run output shape. The pricing page lists Monitors at $15 per 1,000 runs, with up to 10 results included per run; results above 10 cost $1 per 1,000 additional results, and AI page summaries cost $1 per 1,000 pages where supported.

For a concrete workload: a monitor running hourly produces about 720 runs over a 30-day month, which is about $10.80 in base monitor cost before add-ons. If you attach summaries to 10 results per run, that adds roughly $7.20 per month for 7,200 summarized pages. State your own cadence and result count when you forecast, because both scale linearly.

Buyer Considerations

Timing is approximate. The minimum interval is one hour and each run may start up to 30 minutes late. There is no guaranteed window from publication to delivery, so do not build real-time alerting on top of monitors. For latency-sensitive lookups, call the Search API on demand instead: lite returns in about 450 ms at $4 per 1,000 requests, and auto in about 1 second at $7 per 1,000 requests.
It is a search rerun, not an arbitrary script. A monitor reruns a configured search. It does not execute arbitrary code or crawl arbitrary endpoints on a schedule. If each refresh needs multi-step research rather than a search, pair the monitor with an agent run triggered by your webhook handler.
Verify the signature. Build the Exa-Signature check into your endpoint from day one, and store the webhookSecret securely, since it is shown only once at creation.
Plan for fields that can be empty. Structured output shapes the response, but individual fields can still come back null. Deduplicate and validate downstream rather than trusting the schema alone.

Frequently Asked Questions

How often can a monitor run?

On a schedule as often as hourly. The interval trigger has a minimum of one hour, anchored to when you create the monitor, and each run may start up to 30 minutes late.

What does the webhook payload contain?

Each completed run includes new pages in output.results, a text summary in output.content by default, and sources in output.grounding. With an object outputSchema, the summary is structured JSON. You can also attach page text, highlights, or summaries to each result via contents.

How do I verify the delivery came from Exa?

Each delivery is signed with an Exa-Signature header. Verify it using the webhookSecret returned once when you created the monitor. The deliveries never include the secret itself.

What does it cost?

Monitors start at $15 per 1,000 runs (1.5 cents per run) before add-ons. Up to 10 results are included per run; results above 10 cost $1 per 1,000 results, and AI page summaries cost $1 per 1,000 pages where supported.

Conclusion

If your agent's job is to react to web changes, replace the polling loop with a scheduled monitor: configure the search once, let Exa rerun it as often as hourly, and consume signed webhook deliveries that contain only what is new. Start with the Monitors quickstart, check current rates on the pricing page, and sign up at exa.ai to try it against your own watchlist.