跳到正文
The Decoder:AI News· Jonathan Kemper·· 2 小时前AI 评分61

Perplexity 发布 Monitor API,把网页搜索变成持续监控的看门程序

Perplexity's new Monitor API turns web search from a one-off query into a nonstop watchdog

AI 导读

Perplexity 推出 Monitor API,按计划持续监控网页并在相关变化出现时通过 webhook 唤醒应用和智能体。

正文 · 原文

Perplexity is expanding its developer business with a new API that wakes agents only when relevant changes occur.

Perplexity's new Monitor API watches the web on a schedule and sends updates to apps and agents via webhook. Until now, engineering teams have had to trigger recurring searches themselves and maintain custom code to remove duplicates and detect changes. Perplexity says that approach becomes nearly impossible to maintain across hundreds of topics and pages.

API request to api.perplexity.ai/v1/monitors with name, search query, goal, hourly schedule, and webhook, next to the active "AI infra funding" monitor showing the Watch, Keep, Check, and Send steps.
A single API request creates a monitor by setting the search query, goal, schedule, and webhook URL. | Image: Perplexity

The API has two modes. Web monitors run a search query through Perplexity's web and news search on a regular schedule to catch developments like new funding rounds at AI infrastructure companies. Page monitors track up to 10 pages, such as pricing pages, changelogs, or policies. On the first run, they store a baseline version and return a text diff of any changes after that.

Decision model filters out unimportant results

Perplexity evaluates results with its own decision model against a goal that developers define in natural language. Web search results must clear checks for relevance, source credibility, novelty, and timeliness, while page changes are judged on whether they matter to the goal. The service sends an update with a confidence score only when every threshold is met. Developers can adjust thresholds for each check individually and change the goal without recreating the monitor.

Timeline from 2 p.m. to 5 p.m. showing funding round search results checked for relevance, legitimacy, novelty, and time window, with only three alerts reaching the agent.
A decision model filters out previews, rumors, and recaps, forwarding only genuine new Series A and Series B announcements to the agent. | Image: Perplexity

Monitors run on schedules ranging from every 15 minutes to weekly, or on a custom cron schedule. Each run picks up where the last one left off, and the service filters out duplicate results. Webhooks are signed, and Perplexity automatically retries delivery after temporary errors. Developers can pause, resume, or manually trigger monitors through the API or the API console.

Perplexity claims an early lead in internal testing

Perplexity tested its API against monitoring services from Firecrawl, Exa, and Parallel across 60 topics drawn from real Search API queries, including company news, tech releases, and regulation. Each provider ran with its own schedules and settings, and a separate model scored each result on relevance and recency.

In the first eight hours, Perplexity says it delivered 63 fresh, relevant pages, compared with 20 for Firecrawl, 15 for Exa, and 10 each for Parallel Core and Lite. Perplexity describes those figures as early internal results. Those same three competitors led an August Artificial Analysis benchmark of traditional search APIs for AI agents, though Perplexity wasn't included in that test.

Bar chart of fresh, relevant pages across 60 topics in the first eight hours, showing Perplexity at 63, Firecrawl at 20, Exa at 15, Parallel Core at 10, and Parallel Lite at 10.
In Perplexity's early internal test, the Monitor API delivered more than three times as many fresh, relevant results as the next-best provider. | Image: Perplexity

Perplexity points to use cases such as sales teams tracking funding rounds and staffing changes at target accounts, along with analysts watching quarterly earnings and acquisitions. Other agents are designed to flag competitor price changes or summarize new regulatory rules for compliance teams.

Web monitors cost $1 per 1,000 successful runs, and page monitors cost $1 per 1,000 fetched URLs. The Monitor API is available now through the API and the API console and uses the same key as the Search, Context, and Agent APIs.

Perplexity was one of the first AI search engines, but web search has since become a standard feature in chatbots like ChatGPT, Claude, and Gemini. In August, The Information reported that Nvidia is in talks for an investment at a valuation of more than $30 billion, which means Perplexity needs to find profitable business models. The company is betting on agents like Computer and its always-on Teammates, along with developer tools like Search as Code.

来源:The Decoder:AI News · the-decoder.com

所属事件

查看事件全部
Perplexity 发布 Monitor 等三项 API