Living Code OS is made by Duval Software. It began as the team's own company brain and operating board, and it is still the system the team runs on.
Before it was a product, it was the thing we needed. Every meeting recording, mail thread, shared file, call and text went into one memory. A resident agent read that memory, steered the board, and filed anything with a real-world effect for a person to approve. That is the shape of the product today, because it is the shape that worked for us.
We describe it the way we describe it to ourselves. The product overview is written for teams, developers and platform partners, and it is verified against the code: the four-leg retrieval and its rank fusion, the 1,100-character chunks, the default fifty-cent daily budget per fleet worker, the twenty-five dollar PIN line, the seven cron triggers. The pages on this site are drawn from that overview and from nothing else. If the code does not do it, we do not say it.
The rest of it is being built in public view. The build log is where that shows up: a changelog of what landed, a demo of it, and a ledger of what it cost to make. All of it runs on Cloudflare. One Worker is the server. D1 holds the index, R2 holds the originals, Vectorize holds the embeddings, Workers AI does embeddings, extraction and transcription, Durable Objects coordinate the board, Workflows run the long jobs, and cron is the heartbeat. There is no build step.
One inbox, read by the people building it. Questions about the product, the architecture, or what you would point it at go here. No drip campaign follows.
lucas@duvalsoftware.comJeffery is this deployment's resident agent; every deployment names its own. The window on the home page is live: he remembers the product overview and nothing else, and he says plainly when he does not know.
Ask Jeffery on the home page →When there is something to hand you, the early-access list knows before anyone.
Get early access