Perplexity Agent API

Overview

Freshness note: AI products change rapidly. This profile is a point-in-time snapshot last verified on May 24, 2026.

Perplexity Agent API is Perplexity’s managed runtime for building agentic workflows on top of search, browsing, code execution, and sandboxed tool use. It is not just a “send prompt, get answer” endpoint. The product is explicitly positioned as infrastructure for agents that need to think, search, act, and return structured results in one managed surface.

The May 2026 addition to watch is Finance Search. Perplexity now lists finance_search as an Agent API tool for structured public-company and instrument data, including quotes, financial statements, earnings material, analyst estimates, ETF constituents, ownership, and corporate actions.

Key Features

The biggest feature is scope. Perplexity’s current quickstart and changelog frame Agent API as a runtime that can orchestrate multiple kinds of work: grounded search, browser interaction, code execution, sandbox-style task handling, and now finance-specific retrieval. That matters because teams evaluating agent platforms are usually not choosing between one LLM and another. They are choosing whether the runtime can carry the whole workflow.

Perplexity also emphasizes multi-provider support. In the current docs, supported model costs are passed through at direct provider pricing, which makes Agent API feel more like a runtime layer than a model silo. That gives it a practical role for teams that want Perplexity’s search and execution infrastructure without committing every part of the stack to a single foundation model vendor.

Finance Search is useful because it moves some market-data work out of brittle browser scraping. The model can choose fields based on the prompt, so a single agent call can combine valuation context, earnings details, estimates, and recent market data when the workflow needs it.

Strengths

The strongest advantage is that search and execution are packaged together. For research-to-action workflows, the handoff between evidence gathering and task execution is often where systems become brittle. Agent API reduces some of that integration burden by treating search, browser work, sandbox execution, and now finance retrieval as first-class runtime behaviors rather than forcing teams to wire them all up manually.

It also gives Perplexity a more serious enterprise and developer story than the consumer-facing Computer and Comet products alone. If the workflow needs to live inside an internal application or automation layer, this is the more relevant Perplexity surface.

Limitations

This is still a moving product surface. Tool behavior, billing details, and supported runtime patterns can change faster than on older API products. Teams should assume they need explicit evaluation, fallback planning, and budget monitoring rather than treating the runtime as fully mature infrastructure from day one.

Pricing clarity also requires careful reading. The docs are better than marketing-language summaries, but the total spend still depends on model choice plus tool usage rather than a single predictable flat rate.

Finance workflows add a separate correctness risk. Structured data is better than ad hoc scraping, but financial outputs still need source checks, timestamp awareness, and human review before any high-stakes decision.

Practical Tips

Start with workflows where grounded search and bounded action are both necessary, such as market intelligence, browser-based evidence capture, financial research briefs, or internal research assistants that prepare structured outputs. Keep writes human-approved at first, and log every tool call category separately so you can understand what is driving cost and failure rates.

If your need is mainly a human using a browser agent, start with Perplexity Computer. If you need a developer-controlled runtime inside a broader system, Agent API is the better fit.

Verdict

Perplexity Agent API is one of the more interesting runtime surfaces in Perplexity’s stack because it connects search and action in a way that developers can actually build on. It is most useful for teams that want grounded agent workflows without assembling every runtime piece themselves, and the new Finance Search tool makes it more relevant for structured market and company research.