Retriever AI

Retriever AI, branded “rtrvr,” is a browser agent operating from a Chrome extension, hosted cloud browser, API, MCP connection, CLI, or SDK. The same agent can extract data, fill forms, monitor pages, enrich leads, and schedule workflows. The vendor reports 35,000+ customers and 7 million+ tasks, although these are first-party claims.

The multi-surface design is its distinction. Browser Use offers an open-source Python harness plus managed options. Browserbase focuses on programmable browser infrastructure, debugging, proxies, and sessions. n8n is a broader orchestrator. The business automation guide helps select measurable processes before an agent clicks around.

Pricing is transparent. Free Mode costs $0, shows a sponsored result card, applies fair-use limits, and includes 250 welcome credits without a card. Starter is $9.99 monthly for 1,000 credits. Pro is $29.99 for 3,000 credits and one workspace seat; extra seats cost $9.99. Team is $99.99 for 10,000 credits, three seats, and Private Mode with zero workflow data retention. Scale is $499.99 for 50,000 credits, ten seats, Private Mode, and hands-on support. Benchmark credit use to learn effective cost per completed task.

A good first project extracts 100 representative pages into a defined JSON schema. Measure completion rate, duplicates, evidence, credits, retries, time, and corrections. Then test one authenticated Chrome workflow, such as reading an approved account and preparing—but not submitting—a form update.

Advantages are flexibility and continuity: start on an open Chrome page, scale through cloud browsers, and expose the same capability through MCP or API. Published limits make benchmarking possible.

Risks are inherent to browser action. Sites change, CAPTCHAs interrupt runs, and pages can contain prompt injection. Automation can contact the wrong lead, submit twice, publish bad content, or leak data. Use domain allowlists, minimum permissions, schemas, idempotency, spend limits, and approval before messages, purchases, posts, or destructive actions.

Test expired login, changed selectors, ambiguity, slow pages, pop-ups, injection, and partial completion. Verify logs, screenshots, retention, and cancellation. Retriever fits browser work lacking a stable API. If a reliable API exists, direct integration is usually cheaper, faster, and easier to test.

Before scaling, calculate effective cost as monthly plan price divided by verified completed tasks, not attempted runs. For example, a 1,000-credit Starter allowance is economical only if representative tasks consume predictable credits and seldom require retries. Establish concurrency limits and a stop condition for repeated failures. Where a site prohibits automation or handles sensitive personal data, obtain legal and security approval first. Keep screenshots and structured evidence long enough to review failures, while limiting retention to what the workflow genuinely requires.