Products

Perplexity 在 Context API 中集成 Decider,为智能体提供安全抓取

Why we picked it

Perplexity 昨天刚发 Context API,今天就根据 Tobi 的建议加上了 Decider:对抓回的每个网页打 0 到 1 的分,智能体读之前先查有没有提示词注入。自己跑智能体抓网页的人可以直接用。

Perplexity 于 Context API 发布次日为其新增 Decider 集成,起因是 Shopify CEO Tobi Lütke 建议用决策模型来过滤提示词注入等风险。Decider 让开发者对 Context 抓取的每个网页提出 yes/no 问题,模型给出 0 到 1 的分数,智能体可在读取页面前先判断该页面是否含有提示词注入或其他恶意内容。这一流程相当于为智能体提供带自定义评分标准的安全抓取(safe fetch)能力。

Body

技术上,Decider 以打分而非硬性拦截的方式运行,开发者可以用自定义标准来定义何为“可疑页面”,再由智能体根据分数决定是否读取。这种设计把安全判断嵌入了检索管道本身,而不是事后清理。对于运行长时间自主智能体的团队来说,外部网页是不可控输入源,提前过滤可显著降低被注入恶意指令的概率。

Perplexity 的 Context API 定位于为智能体提供网页上下文,Decider 补上了其中缺失的安全层。开发者如果已经在用 Context API 做抓取,只需调用 Decider 对页面提问即可获得分数,无需额外搭建独立的过滤服务。

Original · Aravind Srinivas

We launched the Context API yesterday and a feedback given to us by @tobi was to have decision models check for prompt injection and other forms of filtering. Basically a safe fetch for agents with scores based on custom criteria. We’ve shipped that today by integrating Perplexity Decider with our Context API. Perplexity Developers @perplexitydevs We've added Decider to the Context API. Ask yes/no questions about the pages Context fetches, and Decider scores each one from 0 to 1. Use it to check pages for prompt injection before your agent reads them. 🔗 View Quoted Tweet📊 21 ⚡

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Products" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Perplexity, Context API, Decider, agents.

Source attribution: curated from Aravind Srinivas (@AravSrinivas).