Talorys deploys via npx create-talorys@latest to Cloudflare Pages plus private Worker with SQLite Durable Object
Talorys packages a personal AI agent into Cloudflare's free tier with a single npx command. It uses private Durable Objects for memory and scheduling while staying within documented allocations. The pattern demonstrates verifiable self-hosting without developer-operated infrastructure.
The installer creates a Pages project and a workers_dev:false Worker containing a TalorysAgent Durable Object. All state resides in the object's SQLite tables for conversations, memories, tasks and alarms. Service bindings route API calls without exposing a public URL for the agent. Authentication occurs inside the Worker via a PBKDF2-hashed owner password stored as a secret.
Cloudflare free-tier allocations supply 100000 Workers AI requests daily and Durable Object storage at no cost. The chosen model supports streaming tool calls. No R2, D1 or Vectorize resources are provisioned, keeping the deployment inside documented free limits. Interrupted installs reconcile existing resources without data loss.
Similar patterns appear in Cloudflare's own Agents SDK examples and earlier single-tenant Durable Object demos released in 2024. Talorys differs by enforcing single-user design and by shipping a complete CRUD UI plus scheduled automations rather than raw primitives. Operational risk centers on free-tier AI quota exhaustion and the absence of multi-region replication.
Next deployments will test whether Durable Object alarm scheduling sustains recurring reminders at scale under the 15-minute CPU limit per invocation.
Talorys: 1200 public GitHub stars reached within 60 days of release
Sources (3)
- [1]Primary Source(https://github.com/rociiu/talorys)
- [2]Supporting Source(https://developers.cloudflare.com/durable-objects/)
- [3]Supporting Source(https://developers.cloudflare.com/workers-ai/)