It began as an internally developed tool because my team and I were so sick of having to constantly bounce context around between ai chats, coding agents engines terminals, browsers files scripts,

TurnPanel is a local-first workspace in which an agent can operate throughout your computer, orchestrate other agents and tools, and preserve valuable context over time. Its background maintenance processes transform commonly-used knowledge and tasks into reusable memories and skills stored locally - that's what I mean by self-evolving: it becomes more helpful through tasks finished, but cannot change its own permissions.

It further includes a visible browser with DOM and CDP access, and it does a benchmark of your machine to suggest and set up appropriate local models. You may alternatively connect to codex openrouter opencode or another openai compatible provider, change models between different jobs, and run different models for helper agents.

It is capable of monitoring conversational harnesses like kod and codex, keeping their terminals displayed, and responding to simple queries based on pre-existing contextual information in the workspace.

Also, for constructing persistent dashboard track review queues, and other small utilities, the mini-app SDK is available that eliminate the need of creating a new project and deploying stack. Spaces form a safety boundary by taking care of what memories files services, browser sessions, models, and actions are accessible to each job, and agents are not able to add capacity to the grants. Logins payment destructive actions, and follow-up confirmations hand control back to the user. Complete workspace like jobs terminals browser streams, applications, and long-run processes can also be used via browser or PWA remotely.

Its in public beta for mac and PC, with linux in the works. The basic local version is free, pro adds remote access and higher hosted-service limits. I'll be available to answer questions on how it works.