The Fly Blog

Introduces Sprites as disposable cloud computers for running agents with instant startup and durable filesystems.

Describes the new writable VFS feature for Litestream, an open-source backup/restore system for SQLite.

Explains the design and implementation of Sprites, a new Docker-like technology for running containers without traditional Docker.

Argues that ephemeral sandboxes for agent isolation are obsolete and introduces a new persistent sandbox approach.

Details the Litestream VFS feature for SQLite backup and restore, using a sandwich ratings database as an example.

Encourages readers to build their own AI agents by comparing the learning process to riding a bicycle.

Describes the challenges of gluing servers and network together in Fly.io's platform that transforms Docker containers into micro-VMs.

Recounts a security incident where Fly.io's Twitter account was compromised and explains how it happened.

Announces the release of Litestream v0.5.0, an open-source backup/restore system for SQLite-based applications.

Introduces MorphLLM for building better AI agents, drawing an analogy with audiophile equipment.

Discusses trust calibration for AI software, aligning user trust with actual product capabilities.

Proposes using games as a method for evaluating AI models, with a one-click deploy for AI Town on Fly.io.

Argues that AI projects should focus on a single model rather than being model agnostic for better results.

Presents a new remote AI runtime for Phoenix framework, enabling LLM agents to work with Elixir.

Explains what MCP Servers are and how they integrate with AI clients like Claude and VSCode on Fly.io.

A provocative opinion piece defending AI-assisted programming against skeptical friends who compare it to NFT mania.

Shares practical experience using Kamal 2.0 in production for deploying Rails 8 applications.

Describes a gnarly bug in Fly.io's Anycast router, the largest Rust project in their codebase.

Announces a major revamp of Litestream, an open-source tool for reliable SQLite recovery from object storage.

Launches MCP Servers on Fly.io, discussing the six-month-old protocol's rapid adoption and evolving identity.

Introduces using MCPs (Model Context Protocols) to provision cloud resources like Fly volumes via LLM tools.

Describes building a basic MCP server for flyctl in 30 minutes to enable LLM-driven infrastructure management.

Discusses how automated agents and robots are increasingly becoming the primary users of Fly.io's cloud platform.

Provides a deep technical dive into how Fly.io implements and operationalizes Macaroon security tokens for authentication.

Details a bug and performance optimization journey involving a Rust proxy in Fly.io's global networking stack.

Reevaluates Fly.io's earlier assumptions about GPU demand and shares progress on GPU-enabling their cloud platform.

An exit interview with engineer JP Phillips reflecting on his four years building Fly.io's orchestration system.

Explores how Semgrep's capabilities become more interesting when combined with LLM-driven development agents.

Announces a shift in Fly.io's blogging strategy toward more varied and less formal content.

Critiques VSCode's SSH agent implementation and explores integrating Fly.io with remote editing workflows.