wtf is pi-manager?

Pi Manager is a local web dashboard for people who use Pi Coding Agent, a command line coding assistant. Instead of digging through config files and terminal logs, you open a page at localhost:3001 and get an overview of everything the agent is doing. The dashboard shows your default AI model, how much you have spent today, cache hit rates, and how large your saved sessions have grown, all in one screen. A live status view infers what the agent is currently doing, working, thinking, or finished, by reading its session logs and checking running processes. You can manage the AI model providers you have configured, pull models from a remote source, and set defaults. Sub-agents, which are specialized helper configurations stored as markdown files, can be edited directly from the interface, along with their tool permissions and workflow rules. Other sections let you search and clean up old sessions with a recycle bin for recovery, track token usage and cost trends over time, and browse installed plugins with a whitelist for adding new ones. You can map a local folder as a workspace so the agent can read memory and knowledge files from it. There is also a built in tutorial section with setup instructions and example agents, plus a sidebar chat assistant that asks for confirmation before making any changes. Pi Manager optionally integrates with OpenVL, a separate vision recognition tool made by the same author, which can be installed with one click from the tutorial page if you want the agent to read images. The project has zero production dependencies beyond Node.js version 18 or later, and ships as static files with no build step. It is designed to run locally only. The README warns against exposing the port to the public internet, and notes that exported configuration files have API keys removed automatically. Any settings the dashboard writes are saved with automatic backups of the previous version. It is released under the MIT license.