Exa launches Monitors v2: scheduled web search that alerts your agent only when there's something new, 7 times cheaper

Exa, the AI-focused search engine, has unveiled Monitors v2, a new version of its monitoring product. The company sums up the core idea as “scheduled web search”: you describe what you want to track, and Exa delivers only what is new to your agent, at the cadence you choose. The announcement is dated October 29, 2026, and signed by “The Exa Team.”

Exa pitches the product as an alternative to Google Alerts. According to the post, Google Alerts was built for the inbox. It delivers results by email (usually once a day) or via RSS, as a digest of titles, snippets and links, but it offers no API or webhook that an agent or automation can act on. Where Google sends an email, Exa Monitors calls your agent directly.

The post argues that watching the web for new developments is exactly the kind of task agents should take on. Today, though, most agents handle it like a cron job, running a search every few hours. Nearly all of those searches turn up nothing new, yet the agent still pays, both for the search and for the token cost of filtering new items out of the results. Exa adds that most commercial agents limit users to between 3 and 15 scheduled tasks.

Monitors covers Exa's entire index, which the company says includes 211 million companies, 3.2 million podcast episodes, 1.27 billion product listings and more than 130 billion web pages. It alerts the agent only when something has actually changed. Each run remembers what it has already seen and already delivered. A new page is sent only if it is relevant, comes from a legitimate source and adds something new. A new article about an old story with no additional information does not pass the filter. Results are pushed to a webhook, so the agent never has to poll.

The headline change in v2 is the price. Exa is cutting it to $2 per 1,000 runs, down from $15, making it more than 7 times cheaper. Existing monitor users do not need to change their implementation.

Each run has three stages: - Retrieve: a fast search across Exa's continuously updated index, with a time window matched to the monitor's cadence. - Judge: pages the monitor has already seen are discarded, and the remaining ones must pass four checks: relevance, source legitimacy, new information and event timing. - Compose: pages that pass are turned into a short summary or a JSON object matching the user's outputSchema, with sources and a confidence level for each field.

Exa says that when it ran v1 and v2 side by side on real production monitors, every update v2 delivered was relevant, and monitors became noticeably quieter. This is the company's own claim, and the post gives no detailed figures.

According to the post, an agent can track: - Products, prices and stock: for example, discounts on a specific pair of sneakers or cheap flights from San Francisco to Maui, with an alert when the price changes or an item is back in stock. - News and topics: following a story without rereading the same coverage. - Competitors: launches, price changes and mentions, including those buried in sources that alerts usually miss. These include the index's more than 3 million full podcast transcripts, which make it possible to pinpoint the exact moment, such as minute 34 of an episode, when a competitor was mentioned. - People: executive job changes, or groups of engineers moving between companies.

Exa lists several use cases among teams already building with Monitors: - Prediction market resolution: gathering cited evidence with each update for questions such as how many U.S. universities will announce a formal closure before December 31, 2027. - Portfolio intelligence: investment firms tracking partnership announcements across all their portfolio companies. - Sales triggers: sales platforms catching account signals while they are still news, so teams can prepare outreach. - Adverse media: risk and compliance teams screening counterparties for negative news, lawsuits, regulatory issues, fraud or scandals in order to update risk scores, alert account managers and adjust onboarding plans.

To get started, the post shows a Python example using the exa_py library. A monitor is created with a name, a search query, an interval trigger (for example, a one-day period) and a webhook with the monitor.run.completed event. Each completed run sends the webhook the new pages, a summary or structured output, and citations. Metadata can be added to route each update to the right user or workspace, which is useful when a single agent manages monitors for many people.

In short, Exa turns web monitoring into a push service built for agents. Deduplication and quality filtering happen on the provider's side, and the price has been cut sharply. The performance and index-size figures come from the company itself, and the post provides no independent metrics.