Talorys: Run a Private AI Agent on Your Cloudflare Account

Open-source personal assistant with zero third-party data sharing

GeoDetect PipelineOctober 10, 2026
Share

Overview

Talorys has emerged as a notable option in the growing landscape of private, self-hosted artificial intelligence tools. Unlike many AI assistants that rely on third-party cloud infrastructure or proprietary servers, Talorys runs entirely within a user’s own Cloudflare account. This design prioritizes data sovereignty, ensuring that personal conversations, memories, and task management remain strictly local to the owner’s environment.

The project is free and open-source, positioning itself as a personal AI agent that combines chat capabilities with robust organizational features. It operates without requiring any servers, databases, or accounts managed by the Talorys developers. Instead, it leverages Cloudflare’s existing ecosystem to provide a fully functional assistant that is both private and performant.

Core Features and Capabilities

Talorys distinguishes itself through a suite of tools designed for individual productivity and privacy. At its heart is a chat interface that supports streaming responses, Markdown formatting, and real-time tool activity indicators, offering a seamless conversational experience similar to mainstream commercial AI tools.

Beyond simple dialogue, the agent maintains durable personal memory. Users can view, edit, and delete facts and preferences, which the system selectively recalls during conversations to maintain context without overwhelming the model with irrelevant data. This memory system is integral to the assistant’s ability to provide personalized responses over time.

The platform also offers full CRUD (Create, Read, Update, Delete) management for tasks, notes, and projects. Users can interact with this data either through the dedicated UI or via natural language commands, such as adding a task to review a project the following day. Additionally, an automation engine handles one-time and recurring reminders, daily task digests, and optional AI-driven routines. These automations are powered by Durable Object alarms, ensuring they execute on schedule without keeping any server permanently online.

Technical Architecture and Security

The architecture of Talorys is built entirely on Cloudflare’s serverless platform, utilizing a combination of Pages, Workers, and Durable Objects. The frontend is served via Cloudflare Pages, while a private Worker handles the backend logic using the Hono router. Communication between the browser and the private Worker occurs over a secure service binding, meaning there is no public URL exposed for the agent itself.

Data storage is centralized in a single SQLite-backed Durable Object named personal-agent. This object manages all conversations, memories, tasks, notes, projects, automations, sessions, settings, and usage statistics. The system uses the @cf/zai-org/glm-4.7-flash model via Workers AI for chat inference, supporting streaming and tool calling capabilities.

Security is enforced at the Worker level rather than in the frontend. Authentication relies on a PBKDF2-SHA256 hashed password set by the owner during installation. This password is stored only as a Cloudflare secret, ensuring that plaintext credentials never touch the frontend or external services. The system is designed for single-user access, eliminating the need for complex account management or team permissions.

Deployment and Data Privacy

Deploying Talorys is streamlined through a command-line installer. Users can initiate the process with a single command: npx create-talorys@latest. The installer verifies the Node.js environment (version 20.18 or higher), handles Cloudflare authentication via OAuth or API tokens, and provisions the necessary resources, including the private Worker and Pages project.

The installation process generates unique resource names and a 256-bit session secret. It performs a final verification check to ensure the frontend, authentication endpoint, and storage health are operational before displaying the live URL. Notably, the installer does not run any AI inference during deployment, reducing resource consumption and potential costs.

"All data is stored in a single SQLite-backed Durable Object in your Cloudflare account. Talorys has no telemetry, analytics, tracking, or external data exfiltration."

Talorys explicitly avoids provisioning paid services such as R2, D1, KV, Vectorize, AI Search, or Workflows. It relies solely on the free tiers of Cloudflare Pages, Workers, Durable Objects, and Workers AI. If the free quota is exhausted or Workers AI becomes unavailable, core features like task management, note storage, and reminders continue to function, as they do not depend on AI inference.

The system also includes robust recovery mechanisms. If an installation is interrupted, running the command again will reconcile existing resources without creating duplicates or deleting data. In the event of a forgotten password, users can reset it via the command line, which invalidates all active sessions across devices for security.

This approach offers a compelling alternative for individuals and privacy-conscious organizations seeking an AI assistant that does not compromise data ownership. By keeping the entire stack within the user’s Cloudflare account, Talorys ensures that sensitive information never leaves their controlled environment, aligning with the growing demand for decentralized and private AI solutions.

Share

Sources

Tools mentioned