Pi shipped 34 release events in 30 days, then moved its GitHub and npm identity to Earendil Works. That is the useful signal here. This is not a quiet rename of a dormant CLI, it is a live project changing its distribution path while still publishing at high speed.
ToolVitals scores Pi at 94 overall, with a 94 health score and a 100 shipping score. It also records 30 GitHub releases in 90 days and 48,301 GitHub stars. GitHub currently rounds the repository display to 48.7k stars, but ToolVitals metrics are the source of truth for this post.
The interesting signal is migration under load
Pi describes itself as a minimal terminal coding harness that users can customize with extensions, skills, prompt templates, themes, and packages. The homepage also positions it around multiple execution modes, including interactive use, print and JSON output, RPC, and SDK usage.
That positioning matters because the recent release notes are not just cosmetic churn. Pi 0.73.1 added self-update support for the npm scope migration, interactive OAuth login selection, and JSONC-style parsing for models.json. Pi 0.74.0 then became the first release under the new @earendil-works package scope.
The May 7 announcement says the code moved to earendil-works/pi, npm packages moved to @earendil-works, and the CLI remains pi. It also says older @mariozechner/* packages are deprecated, not unpublished, so pinned installs remain reproducible. That is a clean migration story, not a scorched-earth package rename.
Pi is betting on hackable agent infrastructure
The release stream points in the same direction as the homepage. Pi keeps adding provider support, auth paths, model metadata, extension APIs, and TUI behavior fixes. Recent notes mention Cloudflare AI Gateway, Cloudflare Workers AI, Moonshot AI, Xiaomi MiMo, model thinking-level metadata, OpenAI Codex transport fixes, and session or context handling changes.
That suggests Pi is less interested in being a sealed coding assistant and more interested in being an agent workbench. The official site says Pi skips built-in features like sub-agents and plan mode, and encourages users to build or install the workflow they want. That is a strong product opinion.
The risk is also obvious. Extensibility attracts power users, but it also creates more surface area for breakage. The release notes show lots of fixes around providers, update flows, terminal behavior, exports, and platform-specific edge cases. Fast shipping is good, but agent CLIs live or die on boring reliability.
What ToolVitals cannot infer
ToolVitals can see public signs of life: stars, release events, release frequency, SSL and uptime signals, and score trends. It cannot judge whether Pi’s agent loop is better than a competing CLI, whether the TUI feels good after a week of use, or whether the extension model produces maintainable workflows.
ToolVitals also does not have 30-day commit count or active contributor count for Pi in this payload. That means the release and star signals are strong, but the contributor picture is incomplete here.
The data confidence is 77. Treat the shipping signal as credible, not omniscient.
Compared with nearby tools
Pi is not the hottest tool in this comparison set. n8n has a 240.0 hot score, 187,504 stars, and 51 release events in 30 days. LangChain also has a 240.0 hot score, 136,475 stars, and 32 release events in 30 days.
Pi’s 209.7 hot score is lower, and its 48,301 stars are smaller than LangChain’s count. But Pi’s 34 release events in 30 days beat LangChain’s 32 in this payload. Against Gemini CLI, Pi trails on stars, 48,301 versus 103,818, but ships more frequently, 34 release events versus 17.
That makes Pi a different kind of signal. It is not winning on raw attention here. It is showing a high shipping cadence while changing project ownership and package scope.
Recommendation
If your team wants a coding agent CLI you can customize deeply, evaluate Pi because its current public work is focused on extension points, provider coverage, update flows, and terminal workflow control.
If you want a polished sealed product with fewer knobs, be cautious. Pi’s own pitch is that you change the harness, not your workflow. That is exactly right for teams that like owning their tools, and exactly wrong for teams that want the tool to stay out of sight.