Parallel Search API

What it does

Parallel Search API takes a natural-language objective and optional keyword queries, then returns ranked web URLs and compressed, query-relevant excerpts. Developers call it as a tool from AI agents or workflows, so the model receives a small set of relevant passages instead of raw search result pages. Excerpts can be requested as dense snippets or as full page content in markdown.

The Search API is used as a web search tool inside agent loops, where agents declare semantic objectives instead of constructing boolean keyword queries. Four modes trade latency and depth: Turbo (~200ms), Fast (~700ms), Basic (~1s) and Advanced (~3s). Source domains can be included or excluded, and live fetch with max_age and timeout settings is available for pages that need fresh crawling. Parallel also offers an MCP server and a CLI for integration.

Parallel Search API suits developers adding web retrieval to AI assistants, research applications and agent tool loops. Create an API key, choose a retrieval mode and set the required result count and source constraints. Monthly free credits can support small workloads, while production usage is metered. Teams should account for mode-dependent cost and latency, retain source links and validate retrieved excerpts before relying on generated answers. Request limits and enterprise controls depend on the account agreement.

What it can do

• Accepts natural-language objectives and optional search queries for semantic retrieval
• Returns ranked URLs with compressed, query-relevant excerpts sized for LLM context
• Offers four search modes (Turbo, Fast, Basic, Advanced) with different latency and excerpt depth
• Supports source domain include and exclude controls
• Provides live fetch with max_age and fetch_timeout settings for premium or JS-heavy content
• Returns LLM-friendly markdown content when full page contents are requested

Works with

• MCP server (web_search tool in the Parallel MCP server)
• Parallel CLI
• TypeScript SDK (parallel-web)
• Python SDK (parallel-web)

Example workflows

Grounding a tool call in a research agent

Illustrative example. Input: an agent objective such as finding recent product releases for a named company, plus two keyword queries. Steps: send the objective and queries to /v1/search with mode set to basic, receive ranked results, pass the returned excerpts into the agent's next reasoning turn. Output: a ranked list of URLs with excerpts the agent can cite in its response.

Fresh retrieval with domain controls

Illustrative example. Input: a research objective plus an include list of domains such as government sites. Steps: set source policy to include only those domains, enable live fetch with max_age in hours and a fetch timeout, then request markdown full content. Output: ranked page content restricted to the chosen domains, with timestamps supporting freshness.

Before you choose

• Mode and result count affect price and the number of requests covered by monthly credits.
• Evaluate retrieved excerpts against the application’s own questions and source needs.
• Use the documented current API version for new integrations and check account limits.

Listed under

• Coding
• MCP and agent tools
• AI agents
• API
• Developers
• Grounding
• RAG
• Research

Get started

Generate an API key on Platform, then call the Search API with cURL, Python or TypeScript.

1. Create an API key on the Parallel developer platform and export it as PARALLEL_API_KEY.
2. Install the Python SDK with pip install parallel-web or the TypeScript SDK with npm install parallel-web, or call the REST endpoint directly with cURL.
3. Send a POST to https://api.parallel.ai/v1/search with an objective, optional search_queries, and a mode, then read the returned results and excerpts.

Turbo and Fast cost $1/1,000 requests; Basic and Advanced cost $5/1,000, including ten results. Additional results cost $1/1,000 results. The account offers $5 monthly free credits, enough for up to 5,000 base-rate requests only at the cheapest modes. Signup rewards can total up to $80 subject to conditions. Enterprise capacity and controls require an agreement.

Source: parallel.ai/pricing, checked 29 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Search API - The Best Web Search for Your AI | Parallel ↗parallel.ai
• Parallel Pricing – Pay-As-You-Go Web Search for AI Agents | Parallel ↗parallel.ai
• Search API Quickstart - Parallel ↗docs.parallel.ai

Company website

parallel.ai: product, pricing and policy pages

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.