Talorys launches a self-hosted personal AI agent deployable on Cloudflare’s free tier

Talorys is a free, open-source personal AI assistant that runs entirely inside a user’s own Cloudflare account. It provides chat with streaming responses, durable memory, task and note management, and scheduled automations such as one-time and recurring reminders. The project emphasizes privacy by avoiding any external telemetry or third-party infrastructure controlled by the Talorys developers.

The agent is built on Cloudflare’s free-tier services: Cloudflare Pages for the frontend and Pages Function, Cloudflare Workers for the private API Worker, Durable Objects with SQLite for data storage and scheduling alarms, and Workers AI with the @cf/zai-org/glm-4.7-flash model for chat and tool calling. Talorys does not require paid services such as R2, D1, KV, Vectorize, AI Search, or Workflows.

Deployment is initiated with a single command: npx create-talorys@latest. The installer verifies Node.js (22+), authenticates the user’s Cloudflare account, sets an owner password (hashed locally with PBKDF2-SHA256), and generates unique resource names. It deploys a private Worker with no public URL, a Pages project with a service binding, and verifies the deployment without running AI inference. The installer also supports non-interactive installs via environment variables.

The architecture isolates all traffic to a user-specific *.pages.dev site, with authentication and authorization handled in the Worker. Chat responses stream as Server-Sent Events end-to-end. The agent’s Durable Object manages SQLite-backed storage for conversations, memories, tasks, notes, projects, automations, sessions, and settings. Workers AI provides the chat model, while Durable Object alarms handle reminders and recurring schedules.

By design, Talorys is single-user and does not provision accounts, teams, or external databases. Core features such as tasks, notes, memories, and reminders continue to function even if Workers AI’s free quota is exhausted or unavailable, ensuring continuity without active AI inference.