Twitter has become the default social media tool for much of the Western world. However, for every vital news update or wifejack meme, you’re just as likely to get pages of degen token pushers and pornbot spam. Perhaps you’d just like to moderate your content beyond what is possible with keyword muting. You can’t very well tell Twitter you’re sick of Marvel movies and rid your timeline of superheroes. You need an agent to make these decisions intelligently. That’s why we built Filter, the LLM-powered Twitter mediator running on Kinode.
The LLM Twitter Filter consists of a Chrome extension, which activates when you visit twitter.com, and a Kinode app. Once enabled through the popup settings, the extension communicates with the Kinode app running on your node.
When you scroll through your Twitter feed, the extension crawls the text of each tweet and sends this content to the Kinode app. The app then queries the OpenAI GPT-4 model (or potentially other LLMs in the future) with a user-defined prompt. This prompt includes rules that you specify, along with the tweets themselves. On the backend, GPT-4 analyzes each tweet and returns a binary response - 0 if the tweet breaks one or more rules, and 1 if it adheres to all the rules.
The extension parses this response and hides the tweets that receive a 0, allowing you to focus on the content that aligns with your preferences. If you're curious about a hidden tweet, simply click on the black area where it previously was in your feed. Alternatively, if you come across a tweet that should be hidden but isn't, you can press the injected "x" button to hide it manually (when local models are better, the ‘x’ can be wired upto save the data with labels for further fine tuning) .
You can see this in action, as well as the app setup process, in the video below:
Filter serves as a proof of concept for Kinode's broader vision - enabling developers to create innovative applications by exploiting the composability of other apps and primitives.
As Kinode evolves, exciting possibilities arise from integrating Filter with other applications on the platform. Developers can create new tools, such as personal AI assistants or content management systems. And the composability works in the other direction, as well: imagine Filter learning from your chat or docs and automatically improving its Twitter filtering rules without further input.
Filter's modular architecture is yet another example in a growing line of apps that demonstrates the flexibility and composability of the Kinode platform. By utilizing existing primitives and combining them in novel ways, developers can create powerful new applications that cater to users' unique needs and preferences. This composability enables developers to focus on their core innovations while leveraging the strengths of existing tools, accelerating the development process and fostering a diverse ecosystem of interoperable applications on the Kinode platform.
The LLM Twitter Filter showcases Kinode's mission to enable composable, user-centric computing experiences. By combining the power of browser extensions, decentralized backends, and state-of-the-art language models, we're providing you with the tools to curate your Twitter feed based on your unique preferences.
We believe this is just the beginning of what's possible when users are given programmatic access to the data streams that matter to them. Stay tuned for more innovative applications as we continue pushing the boundaries of what's achievable with AI, social media, and decentralized technologies.
If you want to develop on Kinode, find our docs here. For support and to hang out with the Kinode community, find us on Discord.