Pi released: a minimalist, extensible agent framework with a unified LLM API and coding agent CLI

Pi is a minimalist, extensible agent framework. Its core packages include @earendil-works/pi-ai (unified multi-vendor LLM API supporting OpenAI, Anthropic, Google, and more), @earendil-works/pi-agent-core (agent runtime with tool calling and state management), @earendil-works/pi-coding-agent (interactive coding agent CLI), and @earendil-works/pi-tui (differential-rendering terminal UI). Users can customize via extensions, skills, prompt templates, and themes, packaged as Pi packages shared over npm or git. It supports four usage modes: interactive, print/JSON, RPC, and TypeScript SDK. Sub-agents and plan mode are skipped by default; it requires Node.js 22.19+, and the install script pins dependencies while npm install does not pin transitive dependencies. Pi has no permission system by default to restrict filesystem, process, network, or credential access; it runs with the permissions of the launching user and process. Teams needing stronger boundaries can use the Gondolin extension (runs built-in tools in a local Linux microVM), plain Docker (isolates the whole process in a container), or OpenShell (policy-controlled sandbox). This directly affects developers using the Pi coding agent CLI, especially teams integrating Pi into automated workflows or running local open-source models. On contributions, new issues and PRs from new contributors are auto-closed by default, with maintainers reviewing daily. The project hardens its npm supply chain: direct external dependencies are pinned to exact versions, .npmrc sets save-exact=true and min-release-age=2, pre-commit hooks block accidental lockfile commits, and CI runs npm ci --ignore-scripts with periodic npm audit.