Perplexity shipped Agent API, a managed runtime that handles the full agentic loop from planning to execution.

The service replaces multiple infrastructure components — model router, search layer, embeddings provider, sandbox service, and monitoring stack — with one endpoint. Developers get integrated search, tool execution, and multi-model orchestration through a single API key.

How the agentic compute model works

Unlike traditional processors that execute predetermined instructions, Agent API uses frontier language models as the processor. The model receives an objective, decomposes it into a plan, selects tools, executes them, evaluates results, and iterates until the goal is met.

For sales call preparation, a single request with CRM, web search, and URL fetch tools lets the model automatically gather internal conversation history, find recent competitive intelligence, and pull detailed content from relevant pages.

The API supports model fallback chains across all frontier providers. If one model becomes unavailable, the system automatically switches to the next, targeting near-100% uptime for production applications.

Built-in and custom tools

Two native tools come standard: web_search and fetch_url. Web search supports domain filtering for up to 20 domains, recency filters, date ranges, language selection, and configurable content budgets per page. URL fetch retrieves and extracts full page content from specific links.

Developers can add custom functions to connect agents to their own databases, APIs, and backend systems.

Optimized presets for common workflows

Perplexity offers pre-configured setups for specific use cases: fast factual lookups, balanced research, deep multi-source analysis, and institutional-grade research. Each preset includes recommended system prompts, tools, and cost profiles.

The advanced-deep-research preset runs the same multi-step reasoning engine that powers Perplexity's consumer product. It performs dozens of searches per query and reads hundreds of source documents.

All preset parameters can be overridden. Developers can use presets as starting points and adjust models, tools, step counts, or token budgets in individual requests.

Agent API is available now with documentation at docs.perplexity.ai.