Keak is a representative example of the 'Specialized Agent' layer in the AI ecosystem. Rather than building a general-purpose assistant, they have developed an agent specifically for the CRO domain. Their 'Keak Agents' product is significant because it is granted write-access to a company's most sensitive digital asset: the production website. This marks a transition from AI being a consultative tool to being an active operator in the business stack.
For developers and users in the agent ecosystem, Keak demonstrates how agents can solve the 'closed-loop' problem. By integrating an LLM-based reasoning engine with a statistics engine (SPRT) and a deployment mechanism (the JS snippet), they have created an agent that can observe results, learn from data, and modify its own environment to achieve a specific KPI—in this case, conversion uplift. They are active in the application layer of the agent stack, pushing forward the idea of autonomous site management.
Keak is a Montreal-based startup founded in 2023 that focuses on automating the conversion rate optimization (CRO) process. The traditional A/B testing workflow is famously high-friction: a marketer identifies a drop-off point, a designer creates variations, a developer implements the code, and a data scientist eventually declares a winner. Keak replaces this sequence with an AI-driven cycle that scans page structure, generates copy and layout alternatives, and manages traffic distribution automatically.
The company divides its offerings into four distinct products, catering to different technical requirements. Keak Code is their developer-centric tool for teams that want to manage experiments directly within their codebase. Keak Extension provides a no-code entry point, allowing users to initiate tests through a browser interface. Keak Acquire is a managed service model focused on scaling conversions at a fixed price. The most significant of these, however, is Keak Agents, which represents the company's move toward fully autonomous web optimization.
A common critique of A/B testing tools is their impact on site speed and Core Web Vitals. Heavy scripts often cause a 'flicker' or layout shift as variants load. Keak addresses this with a lightweight 31KB gzipped script and an SPRT-based (Sequential Probability Ratio Test) statistics engine that runs server-side. This allows variants to be applied in under 100ms, which is typically faster than the human eye can detect, ensuring zero cumulative layout shift (CLS). Their tracking pixel integrates with virtually any stack, including modern frameworks like Next.js and React, as well as no-code platforms like Shopify and Webflow.
Keak enters a market recently disrupted by the sunsetting of Google Optimize. While enterprise incumbents like VWO and AB Tasty have added AI features, Keak is built as an AI-native platform from the ground up. Instead of providing a canvas for humans to build tests, Keak is designed for the software to act as the primary optimizer. This 'hands-free' approach is their core pitch to small and mid-sized businesses that cannot justify a full-time CRO team but need to maximize the ROI of their ad spend.
Based in Montreal and operating as a remote-first team, Keak is backed by Maza VC and SNR VC. The team is small, currently in the 1-10 employee range, and focuses on rapid iteration across the AI stack. They have structured their pricing to be accessible, offering a free tier for early-stage sites and scaling up to custom enterprise plans for high-traffic organizations. Their focus remains on landing page optimization—headlines, CTAs, and layouts—rather than ad creative, effectively acting as the final mile in the performance marketing funnel.
An AI agent that autonomously tests, learns, and updates websites to increase conversion rates.
Keak is hiring.