exa.ai
Search API for AI applications
5.0•21 reviews•799 followers
Search API for AI applications
5.0•21 reviews•799 followers
Find perfect lists of leads, candidates or anything in the web. Websets’ semantic search lets you ask in plain English, it agentically sources results and enriches them with you ask for - like emails, tags, and more - all in the time it takes to brew a coffee.
This is the 3rd launch from exa.ai. View more
Exa Monitors v2
Launched this week
Exa Monitors v2 is web search on a schedule for AI agents. Describe what to watch, pick an interval, and Exa calls your webhook only when something new and relevant turns up, with a summary or JSON output and citations. Per Exa, v2 remembers what it has already sent, is quieter than v1, and costs $2 per 1k runs, down from $15. Watch prices, news, competitors, people and more. Python SDK and API available.
Launch tags:Artificial Intelligence
Launch Team
Render Sandboxes: a computer for every agentGive every agent its own machine. Starts in under a second
Promoted
Exa Monitors v2 is the update to Exa’s scheduled web search for AI agents. It’s now cheaper and quieter.
Monitors is web search on a schedule. You describe what to watch and pick an interval, and Exa sends only new, relevant results to your agent’s webhook, so the agent doesn’t have to poll. Exa compares it to Google Alerts, which sends an email or RSS digest, while Monitors calls your agent directly.
What changed in v2:
Price is $2 per 1k runs, down from $15, which Exa calls “over 7x cheaper”. Existing implementations don’t need changes.
Runs now remember what they’ve already seen and passed on, so your agent is only called when something actually changes.
On the team’s production monitors, Exa says v2 updates were all relevant and the monitors were noticeably quieter than v1. That’s Exa’s own testing.
How it works, per Exa: a fast search over a time window that matches the monitor’s cadence, then a check that drops pages already seen and keeps only those that pass relevance, source legitimacy, new information and event timing. What’s left comes back as a short summary or JSON in your own schema, with sources and a confidence level per field.
Things people watch with it, per Exa: price drops and restocks, news on a story, competitor launches and pricing changes, executive job changes, prediction market settlement, sales triggers and adverse media screening.
It’s built for developers, with an API and a Python SDK (exa.monitors.create), and a quickstart in the docs.
Good fit if you’re building an agent that needs to react to changes on the web, instead of re-searching on a loop and filtering the noise yourself.
Try it: Monitors v2 post · Quickstart
P.S. I hunt the latest and greatest launches in tech, SaaS and AI, follow to be notified → @rohanrecommends
congrats on the launch exa team. i would use this for competitor pricing and landing pages, those changes are easy to miss. does the alert show exactly what changed on the page?
The 'only ping me when something actually changed' part is what makes this usable — every monitor I've set up before turned into noise within a week. At $2/1k runs I'd happily leave a bunch running: App Store guideline pages, Swift evolution proposals, competitor changelogs. It's the stuff I check manually every morning and mostly find unchanged.
Cool launch! Does this deliver results from existing search engine indices or has its own crawling/indexing?
@shrsv pretty sure exa runs its own index, thats kind of their whole thing
wed use this to catch companies posting that theyre hiring mobile devs, thats often right before they start asking agencies for quotes. at $2 per 1k runs you can just leave a bunch of them running
cool launch, congrats!! if a story I'm watching gets a follow-up article with one new detail, does Exa count that as new and ping my agent or treat it as already sent?
I've burned a lot of agent cycles on search again, nothing new. If a monitor stays quiet for two weeks, I can't tell whether nothing actually changed on the web, or the judge discarded what I wanted to see. Is there a per-run log of what it threw out?