This talk introduces Bluesky's AT Protocol, a new decentralized social network architecture, and details how users can create custom content feeds. The speaker, "redsolver," presents Skyfeed, an open-source tool featuring a visual editor and a custom scripting language for building these feeds. The presentation covers Bluesky's layered decentralization (Personal Data Services, Relays, AppView), the transparency of its open data streams, and the technical implementation of custom feeds, including how their definitions are publicly available. The talk also delves into Skyfeed's infrastructure designed for scalability through aggressive caching and an efficient Rust-based query engine, concluding with a live demo of advanced feed creation capabilities and future development plans.
0:14 Introduction to Decentralized Social Networks: The speaker emphasizes the benefits of decentralized social networks like Bluesky over centralized platforms, citing issues such as platform lock-in, data loss during migration, and loss of followers.
1:17 Bluesky's AT Protocol Architecture: Bluesky, launched in late 2022, is a microblogging platform built on the AT Protocol, designed for decentralization by ensuring every system component is easily replaceable. Key layers include Personal Data Services (PDS) for user data hosting, Relays for aggregating data streams, and AppView for providing curated data views.
2:33 Easy Identity Migration: Bluesky allows users to easily change their handles (domain-based identities like @example.com) without losing data or followers, as identity is a separate, persistent layer linked via DNS records.
5:52 Open Data Streams & Jetstream: The AT Protocol makes all public data (posts, likes, blocks) available in real-time via open WebSocket streams from Relays. Jetstream is introduced as a service that simplifies consuming these raw binary streams into more user-friendly JSON events, enabling efficient data consumption with minimal delay (~0.5s).
14:07 Custom Feed Creation: Bluesky uniquely enables users to publish their own custom algorithms, known as "feeds." These feeds are standard JSON records containing metadata and a DID (Domain-based Identity) that points to a feed generator service. Users can then query this service to receive a customized list of posts.
21:53 Introducing Skyfeed: The speaker presents Skyfeed, an open-source visual editor designed to simplify the creation of complex Bluesky feeds for users without extensive technical knowledge. It allows for quick iteration and testing of feed logic.
22:04 Skyfeed's Visual Editor & Features: A live demo showcases Skyfeed's block-based editor, allowing users to define feed logic by combining inputs (e.g., entire network, specific users, hashtags), filters (e.g., regex, content types), and sorting mechanisms (e.g., by like count, randomness).
24:51 Transparent Feed Definitions: A core tenet of Skyfeed is the transparency of feed logic. The definition of how a feed is built is published as part of the feed's AT Protocol record, allowing any user to "view source" and understand its functionality. This enables a "Fork Feed" feature for easy modification.
27:35 Robust Infrastructure & Scalability: Skyfeed's backend utilizes an aggressive caching layer (Dragonfly DB) for fast delivery of non-personalized feeds and an efficient Rust-based query engine for on-demand computation of unique feeds, handling high request volumes.
31:02 Advanced Feed Customization: Skyfeed supports advanced features, including custom scripting for filtering (e.g., by like count) and scoring, personalized feeds (e.g., "Mutuals," "Memory Lane"), and integration with external tools like Skygraph for community detection.
44:11 Significant Adoption & Future Outlook: Out of approximately 100,000 Bluesky feeds, 86,000 are hosted on Skyfeed. The project, currently supported by GitHub Sponsors, is exploring new funding models and technical improvements (e.g., easier self-hosting, donation-based personalized features) to sustain its growth and decentralization efforts.