Talorys – A Self-hosted Personal AI Agent On Cloudflare’s Free Tier

TL;DR

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

Talorys is an open-source personal AI assistant that its GitHub project says can be deployed into an individual’s Cloudflare account using a command-line installer. It stores data in a SQLite-backed Durable Object and uses Cloudflare Workers AI; the project says it is designed for the free plan, but Cloudflare quotas apply and may change.

Talorys, an open-source personal AI assistant, is presented by its GitHub project as a way to run chat, memory, task management and scheduled reminders inside a user’s own Cloudflare account. Its installer, npx create-talorys@latest, deploys the app using Cloudflare Pages, Workers, Durable Objects and Workers AI, with the project saying it is designed for the free plan rather than a server or account run by Talorys developers.

The project says Talorys offers streaming chat with tool activity indicators, editable personal memories, and tools for tasks, notes and projects. Users can ask the assistant to add a task through chat, while reminders, recurring schedules, daily task digests and optional AI routines are delivered through an in-app notification center. Scheduled jobs use Durable Object alarms, according to the project’s description, so a separate server does not need to stay online.

Talorys places its application data in a SQLite-backed Durable Object in the user’s Cloudflare account. Its described architecture routes browser requests through a Pages Function to a private Worker, which then connects to the agent. The project says the Worker is deployed without a public URL and that authentication and authorization are handled in the Worker. Chat uses Cloudflare Workers AI and the model identifier @cf/zai-org/glm-4.7-flash.

The installer checks for Node.js 22 or later, guides the user through Cloudflare authorization and account selection, requests an owner password, and deploys the Worker and Pages site. The project says the password is hashed locally with PBKDF2-SHA256 before being stored as a Cloudflare secret. Its deployment check tests the site, authentication endpoint and storage health without running AI inference. A successful setup prints a Cloudflare Pages URL for access.

Personal AI Inside a Cloudflare Account

Talorys brings together features often spread across a hosted assistant and separate productivity apps: chat, memory, task tracking and reminders. Running the application resources in a user’s Cloudflare account gives the account holder control over deployment and stored data, rather than relying on a Talorys-operated application server, according to the project.

That arrangement does not mean the data stays exclusively on the user’s own hardware or is invisible to service providers. Cloudflare provides the hosting and AI inference, and the project says chat messages and relevant memories are sent to Workers AI for responses. The trade-off for readers is between managing an assistant in their own cloud account and accepting the provider’s service terms, account limits and operational requirements.

The project also says core features such as tasks, notes, memories and reminders continue to work when Workers AI is unavailable or its free allocation is exhausted. That separation may make the app usable for basic organization even when AI chat is temporarily unavailable, although the source does not provide independent reliability or performance results.

How the Cloudflare Deployment Works

The documented setup uses Cloudflare Pages for the frontend and a Pages Function for API requests. A service binding forwards those requests to a Worker that the project describes as private; the Worker connects to a Talorys agent built with the Cloudflare Agents SDK. The agent’s Durable Object stores conversations, memories, tasks, notes, projects, sessions, settings, usage data and automation records.

Talorys is described as single-user by design, with no signup, team accounts or Talorys-managed user service. The installer sets an owner password, while users can sign in from multiple devices as the same owner and manage sessions in security settings. The project says a password reset replaces the secret and signs out all devices.

The GitHub material says Talorys does not provision R2, D1, KV, Vectorize, AI Search, Workflows or paid services. It also says the project includes no telemetry, analytics, tracking or advertising code. Those are statements from the project documentation, not an independent audit. Cloudflare still processes the data needed to host the app and run AI inference.

“Your personal AI agent. Your own cloud.”

— Talorys project description on GitHub

Free-Tier Limits and Security Questions

The source material does not provide an independent security review, third-party deployment test, usage benchmark or cost analysis. Its descriptions of private networking, password handling and lack of telemetry are project claims; readers would need to review the linked architecture and security documentation and Cloudflare’s own policies before using the assistant for sensitive information.

“Free” is conditional on Cloudflare’s account-level limits. The project identifies request, Durable Object and daily Workers AI allocations as constraints, and says quotas can change. The supplied description cuts off while explaining what happens after the AI allocation is exhausted, so the exact reset behavior is not fully available in the source excerpt. Actual charges, if any, would depend on the account’s plan, usage and Cloudflare’s current terms.

The source also does not state a release date, adoption figures, or how the app performs under real workloads. It describes semantic recall as optional and off by default, but provides no accuracy evaluation. Those details should not be inferred from the feature list.

Deployment and Cloudflare Quotas

Readers interested in trying Talorys can review the project’s GitHub repository and deployment, architecture and security documents, then run the installer from a local environment with Node.js 22 or later and a Cloudflare login or appropriately scoped API token. The installer is designed to create the resources and report the resulting Pages URL; the project says rerunning it from the same directory reconciles existing resources rather than duplicating them or deleting data.

After deployment, the practical next steps are to check current Cloudflare plan limits, confirm the account and AI settings, and decide what information is appropriate to send to Workers AI. The project’s documentation and Cloudflare’s service terms are the available references for data handling and quota behavior. No separate launch milestone or future product update is specified in the supplied material.

Key Questions

What is Talorys?

Talorys is an open-source personal AI assistant that the project says can be deployed in a user’s Cloudflare account. It combines chat with memory, tasks, notes, projects and scheduled reminders.

Does Talorys run entirely on Cloudflare’s free plan?

The project says it is designed to use Cloudflare free-plan services and does not enable paid features on its own. It is not unlimited free: account-level quotas apply, and Cloudflare can change them. Users should check their current plan and usage.

Where does Talorys store personal data?

According to the project, application data is stored in a SQLite-backed Durable Object in the user’s Cloudflare account. Cloudflare also processes chat messages and relevant memories when Workers AI generates responses.

Does Talorys need a server that stays online?

The project says it does not require a separate always-on server. Its scheduled reminders and recurring automations use Durable Object alarms on Cloudflare.

What happens when the Workers AI free allocation runs out?

The project says chat displays a message when the AI allocation is exhausted, while non-AI features such as tasks, notes, memories and reminders continue working. The supplied documentation excerpt does not fully specify the reset details.

Source: hn