From Voice Agents to Document Intelligence: How AI Startups Run on Cloudflare

AI startups face a familiar set of hurdles: training and deploying models, navigating ethical questions, earning user trust, and keeping up with a fast-moving regulatory environment. But infrastructure reliability is rarely one of those challenges. The companies below built their products on Cloudflare's developer platform, and while their use cases differ widely, they share a common thread: they trust the platform to scale globally without forcing them to manage servers.

Azule: AI Agents for Ecommerce Customer Service

Azule, based in Calgary, Canada, applies AI to ecommerce customer service. The platform lets businesses create AI agents that interact with customers, handling not just support queries but ongoing tasks across industries. The company builds entirely on Cloudflare, with the exception of API calls to OpenAI.

Several Cloudflare products power the application:

  • Durable Objects and WebSockets for live chat
  • Queues for data processing
  • R2 for all data storage, including vector storage—the company uses R2 and the cache API for on-disk vector search instead of a dedicated vector database
  • Workers KV for frequently accessed configuration data
  • D1 for the user database
  • Workers AI (formerly Constellation) for labeling and summarization tasks
  • Workers for Platforms to deploy custom features for users
  • Pages for hosting landing pages and marketing content

The company also uses API Shield, email workers, mail channels for email integration, log push, and outbound workers. Founder Logan Grasby credits Cloudflare's design for speeding up development: "Durable Objects enabled us to build our MVP faster than we could have on any other platform." He also highlights cost savings from moving search functionality to R2. Storing and searching large customer datasets with previous solutions like Pinecone or Milisearch would have cost thousands of dollars per month per customer; with R2 and the cache API, Azule can search large datasets in under 900ms at a fraction of that cost.

42able: Making Business Documents Interactive

42able, headquartered in Wales, UK, builds AI-driven tools to help businesses engage with documents—contracts, reports, invoices, and internal communications. The goal is to turn static documents into dynamic assets that users can query and interact with efficiently.

The company chose Cloudflare for performance compared to other providers, citing no cold boot times, competitive pricing, ease of use, strong local development features, and responsive support. 42able uses Workers for all APIs and recurring cron scripts, Pages for front-end hosting, KV for Angular apps, R2 for cached personal user data, and Cloudflare's DNS management, DDoS protection, Turnstile, and Zero Trust for securing login pages. The team has started testing Workers AI (formerly Constellation) to host models and D1 for database needs.

Co-founder James Finney noted the unique challenges of AI products: "Users expect AI to be perfect or near-perfect, and are much less forgiving of an AI making an error compared to a human making the same mistake." He added that Cloudflare simplifies the training and deployment burden compared to managing individual components.

Layercode: Low-Latency Voice AI

Layercode helps developers build voice AI agents for web, mobile, and phone. The platform provides the real-time infrastructure needed for responsive voice applications at global scale.

Cloudflare Workers power the core of Layercode's voice AI pipelines, delivering zero cold starts for reliable voice data processing. Durable Objects and DO SQL provide the consistent execution and storage environment for hosted voice AI agents. WebRTC support comes through Cloudflare Realtime's product suite, removing the complexity of hosting and scaling WebRTC infrastructure.

"Latency is everything when it comes to natural conversation with voice AI," said Damien Tanner. "Layercode chose Cloudflare because it's the best platform for delivering low-latency voice AI experiences globally."

Eclipse AI: Customer Feedback Analysis at Scale

Eclipse AI, based in Melbourne, Australia, unifies Voice of Customer channels—phone calls, video calls, emails, support tickets, public reviews, and surveys—then analyzes them at scale using generative AI to produce actionable business insights. The company operates a freemium model that aims to make data-driven decision-making accessible to businesses of all sizes.

Eclipse AI runs on a Cloudflare stack: Workers for the backend API, Pages for the frontend across hundreds of cities, Images for serving cascaded asset versions, and R2 as the object store. Founder Saad Irfani noted that switching from S3 to R2 produced "a staggering 70% reduction in cost." He added that Cloudflare's developer platform lets his small team "focus on building our product and not be burdened with managing infrastructure," with automatic scaling under a pay-as-you-go model.

Embley: Marketplace Automation

Embley, based in Sierre, Switzerland, combines AI with process mining to automate and scale marketplace commerce operations. Cloudflare Pages serves the "control center"—the core back-office interface users manage their marketplace operations through—at the edge. Workers power the serverless backend, connected to the frontend through the API Gateway.

Laurent Christen cited three reasons for choosing Cloudflare: "powerful serverless products that enable us to run an entire tech stack without having to care about infrastructure," the scalability of the global network, and security built in through the Zero Trust platform, which protects both production and lower environments, including internal systems and apps.

ChainFuse: Multichannel Feedback Intelligence

ChainFuse, based in San Francisco, helps support, community, and product teams collect and analyze user feedback at scale across multiple channels. Their AI-powered tools surface actionable insights for product decisions.

Workers power the core backend infrastructure, while Cloudflare security solutions protect both the applications and employees. The WAF blocks an average of 48% of all incoming traffic, according to the company. Rate limiting prevents abuse, and API Shield, used alongside the WAF, intercepts an average of 1.32% of traffic that gets past WAF. Zero Trust Gateway not only secures employees but is also integrated into the product itself, preventing end users from exploiting the platform for malicious purposes.

Victor Sanchez said the team has used Google Cloud and AWS, but Cloudflare has stood out: "Since 2016, we have consistently chosen Cloudflare for our projects due to their excellent product range and reliable performance. Saying 'it just works' is an understatement."

ai.moda: Rapid-Fire AI Plugin Development

ai.moda, headquartered in Grand Cayman, is building tools to bridge humans, developers, and machines. Their projects include ChatGPT plugins for CVEs and S3 storage, YourCrowd (an MTurk-compatible API for humans and bots), and Valkyrie, an automated zero-trust hardening tool for Linux applications and cloud workloads.

The company relies heavily on Workers, Email Workers, Pages, R2, Durable Objects, Workers AI (formerly Constellation), Cache API, DMARC management, Access, WAF, logpush, DNS, Health Checks, Zaraz, and D1. The key draw was the Workers platform itself—deploying code globally with a single command simplified DevOps and removed concerns about scaling resources.

David Manouchehri described the speed this enables: "If you want a new ChatGPT plugin, let us know on Friday, and by Monday we can have it developed and shipped in production!"