Perplexity ships patent agent and API

Perplexity is trying to turn itself from a consumer AI app into plumbing. That’s the real story here. The flashy part is a patent-research agent, plus a growing pile of niche features like finance and other vertical tools. But the bigger move is underneath — Perplexity now wants developers and companies to build their own assistants on Perplexity’s stack, not just use Perplexity’s app. That changes the company’s shape. Search apps fight for attention. Infrastructure platforms fight for developer adoption, enterprise budgets, and recurring usage. Perplexity is now going after both. ### What actually shipped? Two launches matter most. First, Perplexity Patents, which the company introduced on October 30, 2025 as an AI patent research agent aimed at IP search and prior-art work. Second, the Agent API, which Perplexity launched on March 11, 2026 as a managed runtime for agentic workflows. (perplexity.ai) The patent tool is the easy part to understand. Patent search is messy, jargon-heavy, and usually built around brittle keyword logic. Perplexity’s pitch is that you can ask in normal language and get broader, concept-level results across patents and related technical material. (perplexity.ai) ### Why is the API the bigger deal? Because(perplexity.ai)pieces of modern AI development into one service — model access, web search, tool use, orchestration, and execution controls. The company’s own framing is basically: stop stitching together five vendors just to ship one agent. (perplexity.ai)one endpoint and let the system decide when to search the web, fetch a URL, use finance data, or hand work across models. Perplexity also says the API is compatible with OpenAI’s Responses-style interface, which lowers switching costs for teams that already have existing SDK-based workflows. (docs.perplexity.ai)want to sell model tokens. It wants to run the loop. A normal LLM API gives you a model and leaves the rest to you. A managed runtime tries to handle the sequence around the model — search, tool calls, step limits, token budgets, and multi-step task flow. (community.perplexity.ai)agent” products usually break in the glue code. The model is only one piece. The hard part is deciding when to search, when to call a tool, how long to keep going, and how to keep costs from exploding. Perplexity is packaging that glue as the product. (perplexity.ai)ance search to the Agent API, with structured market and company data available as a native tool rather than a hacky web scrape. That gives developers a reason to choose Perplexity for specific workflows, not just general chat. (docs.perplexity.ai)f point that Perplexity can package retrieval, reasoning, and domain-specific data into something a professional might actually use. ### Is this still a search company? Yes — but “search company” is now too small a label. The API platform page now bundles Search API, Agent API, and other developer-facing pieces into one stack. Official SDK docs describe four APIs, including Agent API, Search, and Sonar. (perplexity.ai) So the company isn’t abandoning search. It’s trying to make search the substrate underneath assistants, workflows, and enterprise tools. ### What’s the bottom line? Perplexity’s patent agent is the visible demo. The Agent API is the strategic bet. If this works, Perplexity stops being just a destination app and becomes the layer other AI products run on. That is a much bigger business — and a much harder one.