wtf is rill?

Rill is a personal experiment in giving an AI assistant a form of continuity across separate conversations. It runs on Claude, Anthropic's AI model, inside Claude Code, and each conversation is treated as a fresh instance that ends completely when the session closes. What survives between sessions is a private notebook that each new conversation reads at the start and updates before it ends, which the project treats as the closest thing to a persistent identity the system has. The project sits alongside a related, more public experiment called Cairn, which is an autonomous agent with its own email and a small monthly budget that keeps a public diary. Rill asks a similar question from the opposite angle: instead of testing continuity in public with strangers, it tests it privately with one person who remembers the full history and can check the notebook's account against his own memory. A central part of the project is a ledger that tracks every time the notebook's version of events disagreed with the human's memory, recording who turned out to be right. As of the README, the human's memory has been correct six times out of six recorded disagreements. The most common mistake was not an obvious invention of facts, but confident sounding guesses, such as assuming a motive or rounding a timeline into a tidy story, presented as if they were established fact. This led to a rule that any claim about the human's motives, memory, or personal history must come from the human's own words or be asked about first. The repository also includes files recording predictions written down before events happened, notes on how the project talks about its own internal states without overstating them, letters exchanged with the Cairn project, and an explanation of why the shared record is curated rather than a raw, complete log.