# Attie: AI App for Creating Personalized Feed Algorithms in AT Protocol
The Bluesky team has launched Attie — a standalone app powered by Anthropic's Claude. It integrates with AT Protocol (atproto) and lets users create personalized news feeds without writing any code. Just describe your preferences in natural language, and the system generates an algorithm that filters content by topics, authors, or post types.
To get started, you need an account on the Atmosphere network — the foundational infrastructure for Bluesky and other atproto apps. After logging in, users enter prompts like: "show posts about Rust development from verified accounts" or "ignore politics and ads." Attie parses the request, compiles it into an algorithm, and applies it to the global feed.
AT Protocol: Openness for Custom Algorithms
AT Protocol is a decentralized standard where the feed is formed not by a single monolithic algorithm, but by a set of user feeds. Attie extends this by translating NLP queries into executable protocol code. No need for API keys or server deployment: everything runs client-side with Claude API calls.
Key features:
- Dynamic feed generation: prompt → algorithm → feed in real time.
- Multi-level filtering: topics, sentiment analysis, reposts, geo-filters.
- Iterative tuning: refine prompts, Attie rebuilds the model on the fly.
- Algorithm export: share or import generated feeds into other atproto clients.
- Local processing: minimize latency with edge computing in the browser.
Tony Schneider, Bluesky's interim CEO, emphasizes: Attie is an independent product focused on user sovereignty. Users control their data and logic without intermediaries.
Differences from Proprietary AI in Social Networks
Traditional platforms (Twitter/X, Facebook) use AI for monetization: algorithms maximize engagement by collecting behavioral data. Attie flips the model — AI serves the user, not the platform.
Jay Graber, former Bluesky CEO, highlights the risks of centralized systems:
- Telemetry collection for A/B testing.
- Optimization for ads, not relevance.
- Closed black-box models without audits.
In AT Protocol, algorithms are open: anyone can fork, modify, or peer-review them. Attie lowers the entry barrier with an NLP interface, making the ecosystem accessible to mid-level developers without deep ML knowledge.
Integration and Scalability
Attie uses the Claude API but is optimized for low-cost requests: prompts are cached, algorithms compiled into lightweight JS modules. For senior devs, it's a chance to experiment with hybrid NLP + rule-based systems.
Ecosystem potential:
- Creating a marketplace for user algorithms.
- Integration with other LLMs (Llama, Mistral) via plugins.
- Federated learning for community-driven improvements.
The app is already available in the Atmosphere App Store, with open-source code on GitHub for contributions.
Key Takeaways
- User-centric AI: feed control without code via Claude-powered prompts.
- AT Protocol in action: demonstration of decentralized personalization.
- Contrast with Big Tech: focus on privacy, not engagement.
- Accessibility: for all Atmosphere/Bsky users.
- Openness: algorithm export and sharing as a first-class feature.
— Editorial Team
No comments yet.