Continuo

I built Continuo as an AI harness for coding and other work. It centers on a single infinite chat with built-in memory, so I can keep the same conversation going over time. Background threads take on work without blocking the chat.

Memory, Threads and Projects

A model's context has a limit, but a useful assistant needs access to earlier conversations. Continuo keeps the original conversation and builds a navigable tree of summaries. The assistant can zoom into older material when it needs more context. The memory follows Victor Taelin's OptChat recipe.

Threads have their own conversations and memory. I can register projects and start threads that work in each project's folder with its own rules and skills. Threads report back when their work is done, leaving the main chat available.

Built to Be Extended

I wanted Continuo to be hackable, so people can modify it, extend it and build their own assistants or interfaces. Its core SDK provides plugin APIs for the server and the UI. Plugins can add tools and policies on the server as well as commands, keys and views in the terminal.

Most built-in features use those same APIs, including memory, threads, background jobs and parts of the terminal interface. Hosts can choose capabilities, storage and policies without the core hard-coding a workspace or interface.

Continuo builds on Pi Durable. A server owns the shared storage and runs independently of its clients, including the terminal. Policy checks guard tool calls, and the memory system waits for a safe handover before reducing the active context.