Open Source

A video player where replies attach to moments on the timeline, and the conversation grows as a tree.

This is the player you see on the home page and on every episode, released so that anyone can build the same kind of conversation on their own site. As you watch, you see places on the timeline where other people have responded. Follow one and the original pauses while you enter a new branch: a video or audio reply, recorded right there in the player. Replies can have replies. When you’re done exploring, Back returns you to exactly where you left off.

The package is one web component with no framework and no runtime dependencies. It includes the viewer, the in-player recorder, the exploration logic that marks what you have and haven’t seen, a written contract for the server side so it works against any backend, and a small reference server with a runnable demo.

What it deliberately leaves to you: sign-in, membership, storage, and moderation policy. The contract says what the player needs from them; how you provide them is your business.

View on GitHub Read the README

A household AI familiar with persistent graph memory

A brief paper on how Wren is built: a language model answers a conversation, and Wren supplies the memory, context, tools, and rules that make that conversation part of a longer relationship. Her continuity lives in a knowledge graph, so replacing a model or ending a session does not mean starting over.

It covers the memory structure (a curated map and a learned network of associations over the same nodes), how a bounded working set is assembled for each conversation, the three modes — personal, practical, exploratory — that read the same graph differently, self-reflection at session close, and the boundary between private household memory and public activity. It describes an implemented architecture and its current limits, not a benchmark or a claim about machine consciousness.

Read the paper