Palet
A networking protocol for portable, contextualized intelligence. And an app client that's a 'generative browser.'
Demo
About
The problem it solves **What is the problem that you’re solving?** People don't have perfect recall. That's why they resolve to leaving 60 tabs open (so they don't lose them), bookmarking hundreds of Tweets, and highlighting blog articles with Readwise. We're solving that by scraping text and transcribing videos to capture everything you see, hear, and search for—passively as you browse. And then storing that as context so you can recall that blog post you read last week or connect the dots between a video you just watched and a tweet that had to do with it. **Expand on the product that you're building** We're building a search agent (think Perplexity), but instead of searching the World Wide Web, it searches your personal 'Contextual Web.' Starting with what you browse. The early version of the browser is a Chrome extension that scrapes sites as you browse them (or text transcriptions of videos). That makes for structured data which can be used to construct a web or graph of 'contextual objects' of what you browse. These objects can be indexed and loaded into a model’s context window whenever the user asks a question. So if the user asks to “Find that Tim O’Reilly blog I read last week” the model can simply perform a typo-tolerant, semantic string matching search to find the relevant objects and load them into context. And use that to answer the question. These contextual objects, which are just messages, can be stored on separate servers and synced across a network built on a federated architecture (called Open Context). That would guarantee that users can switch between apps while keeping their context across any service that utilizes it. And it also ensures that developers can build personalized, context-aware AI apps without being disadvantaged by monopolized context. We got inspiration from Farcaster and ATProtocol.
How it's made
Challenges we ran into - Issues with getting search to work because of embeddings (see demo video). - We weren't able to finish everything on time to be able to put it together.