Attie: Build Custom Bluesky Feeds with AI in 2026

The promise of decentralized social media wasn't just about data ownership; it was about reclaiming the algorithm. While Bluesky's open protocol (AT Protocol) technically allowed users to build their own feeds from the start, doing so required a level of technical proficiency—Python, server management, and API handling—that most users lacked.

In 2026, the barrier to entry has vanished. Attie has emerged as the standard for non-technical users to build, deploy, and manage custom Bluesky feeds using natural language and semantic filtering. Instead of broad keyword matching, Attie leverages Large Language Models (LLMs) to understand the context of posts, ensuring your feed contains exactly what you want and none of the noise.

Why Keyword Filtering is Obsolescent

Traditional social media filters rely on exact string matching. If you want a feed about "Cloud Computing," you might filter for that phrase. However, you will inevitably miss posts about "serverless architecture," "AWS Lambda," or "distributed systems" unless you manually add hundreds of synonyms. Conversely, you will catch irrelevant posts that happen to use the words in a different context.

Attie operates on semantic embeddings rather than keyword strings. It understands that a post discussing "decentralized identity" is relevant to a "Web3" feed even if the word "Web3" never appears. For professionals using Bluesky as a research tool, this shift from "search" to "understanding" is the primary reason to adopt AI-driven feeds.

Getting Started: Connecting Attie to your Bluesky Account

The setup process is data-sovereign. Attie does not require your master password; instead, it uses App Passwords to interface with your account.

1. Generate an App Password: Go to your Bluesky Settings > Advanced > App Passwords. Label it "Attie" and save the generated code.
2. Authenticate: Log into the Attie dashboard and enter your handle (e.g., user.bsky.social) and the App Password.
3. Define Your Persona: Attie allows you to set a "Librarian Persona." This is a global system prompt that instructs the AI on the tone and quality of content you prefer across all your feeds.

Step-by-Step: Building Your First AI Feed

The core of Attie is the "Instructions" panel. This is where you move beyond hashtags and into logic. For this tutorial, we will build a "High-Signal AI Engineering" feed.

1. Define the Scope

Start by telling Attie what the feed is about. Instead of keywords, use descriptive sentences. Input: "Collect posts discussing practical implementations of LLMs, agentic workflows, and RAG architectures."

2. Set the Exclusion Logic

This is where Attie saves you time. You can exclude content types without needing complex regex. Input: "Exclude any posts that are purely promotional, contain 'thread 1/n' structures unless they provide code snippets, and ignore general AI hype or news about stock prices."

3. Establish the Quality Threshold

By default, Attie looks at the text. By toggling the "Account Relevance" filter, you can instruct the AI to prioritize posts from accounts that have a history of posting technical content. Input: "Prioritize posts that include GitHub links, technical diagrams, or code blocks."

Understanding the Logic Backend

When you submit these instructions, Attie translates them into a set of weights. It evaluates the "Firehose"—the live stream of all Bluesky activity—against your semantic criteria.

Advanced Automation: Chaining Feeds

Once you have mastered individual feeds, you can use Attie’s "Compound Feeds" feature. This allows you to create a "Master Feed" that pulls the top 10% of posts from several specialized feeds.

For a developer, this might mean a feed that combines:

• Top 10% of your "Rust Programming" feed.
• Top 10% of your "Open Source Security" feed.
• Direct mentions of your own projects.

This creates a high-density information stream that effectively acts as a personalized daily briefing.

Pros and Cons of AI-Managed Curation

While the efficiency gains are significant, moving your content consumption to an AI-intermediated model requires an understanding of the trade-offs.

Strategic Implementation for Teams

For companies and technical teams, Attie feeds aren't just social tools; they are competitive intelligence tools.

If your team is working on a specific niche—for example, "privacy-preserving machine learning"—you can build a feed shared across the organization. By subscribing to the same Attie CID (Content Identifier), everyone on the team sees the same curated research, fostering a shared knowledge base without the distraction of a general social media timeline.

How to export and share your feed:

1. In the Attie dashboard, click Publish.
2. Assign a Name and Icon.
3. Attie will provide a Bluesky Feed URI.
4. Copy this URI and share it with your team. They can then "pin" this feed in their own Bluesky interface.

Troubleshooting Common Issues

The most common frustation with AI feeds is "over-filtering." If your feed is empty or moving too slowly, check the following:

• Broaden the Instruction: If you are too specific (e.g., "only posts about Python 3.12.4 performance on ARM chips"), the volume will be naturally low.
• Check the 'Like' Threshold: Attie allows you to filter by engagement. Setting this too high on a niche topic will kill the feed. Set it to 0 initially to see the raw AI classifications.
• Semantic Drift: Language changes. If new terms emerge in your field (e.g., a new library name), occasionally update your instructions to include them, though the LLM should catch these via context.

FAQ

The Next Step

If you are tired of the noise on your "Discover" or "Following" feeds, your first actionable step is to create a "Sanity Feed."

Go to Attie, connect your account, and create a feed with the instruction: "Show me posts from people I follow, but remove all posts regarding political outrage, meta-commentary about the platform itself, and repetitive memes."

Experience the difference of a feed that respects your attention for 24 hours. Once you see the quality of content that remains, you can begin building the more granular, professional feeds that turn Bluesky into a legitimate workstation tool.