Talorys turns your free Cloudflare account into a private AI agent
Deploy a personal AI assistant with memory and task management on Cloudflare's free tier using a single terminal command.
Tools · Source: Hacker News
What happened
Developer Roc Yu released Talorys to the public. It is an open-source personal AI agent that lives entirely inside your own Cloudflare account. The project just hit 695 stars on GitHub and is gaining traction among developers. It gives users a highly private assistant equipped with chat, memory, and task management capabilities. The agent can also run scheduled reminders and automated routines.
Deployment takes exactly one command from your terminal. Running a single npx script automatically provisions the frontend on Cloudflare Pages and the backend API on Cloudflare Workers. It sets up a local SQLite database using Cloudflare Durable Objects to handle state. The system leverages Cloudflare Workers AI to run the glm-4.7-flash model for streaming responses and tool calling.
The entire architecture is built strictly to survive on the Cloudflare free tier. It explicitly avoids provisioning paid Cloudflare services like R2 storage, KV, D1 databases, or Vectorize. If you hit your daily AI compute limits, the chat simply pauses. Your tasks, notes, and reminders keep working perfectly without AI. There is zero telemetry, no tracking code, and no central database controlled by the developers.
Key facts
- 695 — GitHub stars for the Talorys repository
- npx create-talorys@latest — The single command required to deploy the entire application
- @cf/zai-org/glm-4.7-flash — The specific Workers AI model used for chat and tool calling
- Node.js 22+ — The minimum requirement for local development and deployment
Why it matters
Builders now have a proven blueprint for zero-cost, zero-maintenance AI applications. Hosting personal tools usually means managing a virtual private server or paying monthly subscription fees to a SaaS provider. Talorys proves you can cram a full-stack agent with persistent memory and scheduled cron jobs directly into Cloudflare's free edge network. You get the absolute privacy of self-hosting without the traditional infrastructure headache. The data never leaves your personal account.
This project also pushes the limits of what edge computing handles natively today. Cloudflare is no longer just a content delivery network or a simple serverless function host. By combining Durable Objects for state, Alarms for background scheduling, and Workers AI for inference, developers can treat Cloudflare as a complete operating system for autonomous agents. Expect a massive wave of single-command deployable personal tools to follow this exact pattern.
For builders
Study the single-command deployment script
The npx create-talorys installer handles authentication, secret generation, and resource provisioning automatically. Builders should copy this deployment pattern to reduce friction for open-source adoption. Users pay zero setup tax and do not have to manually configure environment variables.
Leverage Durable Objects for local agent memory
Talorys uses a single SQLite-backed Durable Object to store conversations, tasks, and user settings. This avoids complex database setups while keeping state directly attached to the compute layer. It is a perfect, low-latency architecture for single-user applications that need reliable memory.
Build resilient apps that survive API limits
The agent gracefully degrades when the free Cloudflare AI allocation runs out for the day. Tasks, notes, and scheduled reminders continue functioning locally without AI intervention. Engineers must design robust systems that do not completely break when external inference endpoints fail or throttle.
My take
I love seeing developers squeeze every single drop of value out of corporate free tiers. Talorys is an absolute masterclass in edge architecture because it proves you do not need a massive cloud bill or complex infrastructure to run a highly useful AI agent. If you are building personal productivity tools today, this is the exact deployment model you should steal.
Original reporting: Hacker News. This is my rewrite and opinion.