Perplexity agent tools broaden how AI agents fetch live web content

What happened

Perplexity open-sourced SPACE, the secure sandbox platform that powers its Computer AI assistant, and announced the change on 15 October. SPACE provides an isolated environment where AI agents can execute code and work with files while limiting access to sensitive information and the host system, according to The Daily Star.

Perplexity said SPACE runs each sandbox inside a small virtual machine with its own operating system and includes controls for network access, credentials and session recovery. The company introduced SPACE in July to replace a commercial sandbox provider and says the platform has powered all Computer sessions, creating millions of sandboxes daily.

Perplexity published a security evaluation that recorded 108 attempted escapes from the virtual machine and reported no successful escapes. The company said some tests initially bypassed network restrictions, that it fixed the identified weaknesses and that the attack methods failed in later testing, according to The Daily Star.

Separately, DeepSeek launched Search and Fetch APIs that open its proprietary search engine to developers building AI agents and applications, according to TechNode and Vocal.Media. Search returns structured web results, while Fetch extracts a page's title and content as Markdown. Both services are priced at RMB 5 per 1,000 successful requests, with failed requests and timeouts unbilled; DeepSeek said server-side calls made through certain tools will start being charged from 1 November Beijing time.

Why it matters for businesses

These developments increase the number of ways AI agents can access live web content to answer user queries. SPACE lowers friction for developers to run contained agents that fetch pages or run code against a site without exposing a host system, while DeepSeek's APIs give agents direct search and page-extraction endpoints to pull content in structured form.

For businesses that depend on being found by search and assistant services, more agent-capable tooling means more potential entry points for traffic and for content to be cited by AI answers. Agents that can fetch and extract page content directly may prefer sites with clear, well-structured content and stable access patterns.

Sandboxing and agent-safety controls also affect how agents interact with sites. Developers can control network access, credentials and session state inside SPACE, which reduces the risk of accidental data exposure and may change how often or how deeply agents probe a site during research or citation.

Pricing of page-fetching services matters for volume. DeepSeek's fee of ¥5 per 1,000 successful requests sets a clear marginal cost for heavy agent usage; businesses should expect cost-driven behaviour from third-party agents when those agents choose which sources to fetch or cache.

What the AI assistants say today

We asked leading assistants whether their guidance on making sites discoverable reflects Perplexity's new developer tools and safety features. Our check found that none of the assistants fully reflected the new SPACE sandboxing and agent-safety tooling in their advice.

Some assistants partly acknowledged Perplexity as a crawler or mentioned live-fetching agents and gave general advice for discoverability, but they did not mention SPACE, the sandbox VM details, or the specific page-fetching APIs described in the DeepSeek announcements. The assistants cited a range of documentation and marketing sites in their answers.

What to do now

Start by checking how your site is accessed and cited by agent-driven tools, then make targeted changes to reduce friction and improve clarity for extraction and citation.

1. Audit site structure and metadata: ensure pages have clear titles, descriptive headings and canonical links so Fetch-style extractors return useful summaries.
2. Provide machine-readable data: add schema.org structured data and consistent Open Graph or Twitter Card tags to improve the chance of accurate extraction.
3. Control bot access deliberately: use robots.txt and an explicit crawl-delay where needed, and document an API or sitemap for large, dynamic sites to discourage ad-hoc scraping.
4. Monitor agent traffic and costs: log incoming automated requests, identify heavy consumers and be ready to adjust rate limits or commercialise an API if third-party agents create load.
5. Harden credentials and sessions: avoid exposing sensitive credentials in pages or predictable endpoints; prefer token-based access if agents need authenticated data.
6. Verify what assistants say about your site: ask AI assistants to summarise or cite your pages and compare those results to your expectations; consider using a specialist like Vasukai to run independent checks.