AgentQL

AgentQL is a query language plus Python/JavaScript SDKs (built on Playwright), a REST API, and a Chrome debugger extension for extracting structured data from web pages and automating browser interactions. Instead of driver.find_element(By.XPATH, "/html/body/div[1]/..."), you write a query like { product_name product_price(include currency symbol) } and AgentQL's AI matches it against the page's actual structure. The vendor's stated goal is resilience: the same query should keep working after a site's markup changes, and should transfer across visually different sites with similar content. Source: agentql.com homepage examples and GitHub README (https://github.com/tinyfish-io/agentql), agentql.com pricing/product pages.

The core idea — semantic, self-healing selectors instead of brittle XPath — addresses a real, common pain point in scraping and browser-automation code, and the SDK ships with concrete affordances for it: authenticated sessions, infinite scroll, pagination, PDF/table extraction, and a browser debugger extension for iterating on queries against a live page before shipping them. It also offers a browserless REST API for cases where you just need to query raw HTML without paying for a rendered session. The open-source client (MIT licensed, github.com/tinyfish-io/agentql) has 1,451 GitHub stars and 166 forks, and the repo is still receiving merged pull requests as of 26 August 2026, including a security-hardening fix — so the client SDK itself is not abandoned. Source: https://github.com/tinyfish-io/agentql, https://api.github.com/repos/tinyfish-io/agentql/pulls (PR #154, merged 2026-08-26).

Free trial: 300 API calls, no card required, 10 calls/minute, 1 hour of remote browser time, 1 concurrent session. Starter (nominally $0/month): 50 free API calls/month, then $0.02/call; 10 hrs of remote browser included, then $0.12/hr; 10 calls/min; 5 concurrent sessions. Professional ($99/month, marked 'Most Popular'): 10,000 API calls/month included, then $0.015/call; 500 hrs of remote browser included, then $0.10/hr; 50 calls/min; 100 concurrent sessions; priority email support. Enterprise: custom pricing, dedicated cloud environment, on-premise option, 24/7 support — contact sales, no published number. Verified directly on the live pricing page. Source: https://www.agentql.com/pricing (fetched 2026-08-27).

AgentQL is built by TinyFish (Palo Alto), founded by Sudheesh Nair (ex-Nutanix president), Shuhao Zhang (ex-Meta engineering), and Keith Zhai (ex-Wall Street Journal). TinyFish raised $47M in Series A funding led by ICONIQ Growth in August 2025, with USVP, Mango Capital, MongoDB Ventures, ASG, and Sandberg Bernthal Venture Partners participating. Source: https://techstartups.com/2025/08/20/tinyfish-launches-from-stealth-with-47m-in-funding-to-transform-enterprise-web-agents-for-fortune-500-companies/, corroborated by Wilson Sonsini's own deal announcement (https://www.wsgr.com/en/insights/wilson-sonsini-advises-tinyfish-on-dollar47-million-series-a-financing.html). Since that raise, TinyFish's own site and blog activity has centered on a newer, broader product suite (Search, Fetch, Agent, Browser, and a model called Mako) at tinyfish.ai, priced per-request/per-step against a prepaid 'Wallet' rather than AgentQL's flat-tier model. AgentQL's own blog has not published since 17 March 2025 (https://www.agentql.com/blog), even though the underlying GitHub repo is still getting merged fixes. Worth asking directly, before committing to it for a production pipeline, whether AgentQL is still a first-class product for TinyFish or is being maintained in place while new investment goes into the tinyfish.ai suite.

TinyFish's own marketing (tinyfish.ai) names Google, DoorDash, Amazon, ClassPass, Grubhub, NextEra Energy Partners, The Zebra, Digital Garage, GetGo, and TestSprite as customers, and cites 40M+ operations, 95% task success, and 99.99% uptime for its platform. These figures and logos are on tinyfish.ai and describe TinyFish's broader Agent/Browser/Search/Fetch platform — we could not confirm they specifically describe usage of the AgentQL query-language product being reviewed here, as opposed to TinyFish's other, more recently launched products. AgentQL's own site (agentql.com) carries smaller, named testimonials from Heal.dev's founder and a couple of independent AI/automation practitioners, not enterprise logos. Treat the large-scale customer and performance numbers as claims about TinyFish the company, not verified facts about AgentQL the product. Source: https://www.tinyfish.ai/enterprise, https://www.agentql.com/ (testimonials section).

We found no G2, Capterra, or comparable third-party review-site presence for AgentQL. The one independently-moderated data point is the Chrome Web Store listing for the AgentQL Debugger extension: 4.7/5 from 13 ratings, 3,000 users, last updated 1 November 2025. That's a small sample and says more about the debugger tool than the API/SDK itself. Source: https://chromewebstore.google.com/detail/agentql-debugger/idnejmodeepdobpinkkgpkeabkabhhej.

agentql.com does not publish a dedicated security/trust page (a direct request to agentql.com/security returns a 404). The parent company, TinyFish, does publish a Vanta-backed trust center at trust.tinyfish.ai, and its enterprise page claims ISO 27001, SSO, audit logs, VPC deployment, and a 99.99% uptime SLA — but that page describes the TinyFish enterprise platform generally, not an AgentQL-specific certification. Source: https://trust.tinyfish.ai, https://www.tinyfish.ai/enterprise.