pi.el

pi.el is a terminal-free Emacs frontend for the pi coding agent.

It starts hidden pi --mode rpc subprocesses, keeps one session per project or directory scope, and renders conversation output in Emacs side windows. The package is meant to stay lightweight, editor-native, and easy to use from the buffer you are already working in.

Status

Early and evolving. This package targets @earendil-works/pi-coding-agent 0.81.1 or newer and intentionally follows the current RPC protocol without compatibility shims for older pi releases.

Pi RPC mode follows pi's project-trust rules. If you want Emacs sessions to load project-local .pi settings, extensions, skills, prompts, or themes, trust the project from interactive pi first, configure pi's defaultProjectTrust, or add --approve to pi-rpc-extra-args.

Features

• terminal-free RPC integration with pi --mode rpc
• one hidden pi session per project root or directory scope
• side-window session buffer with live streaming output
• prompt composer buffer with editor-native editing, defaulting to org-mode
• @file completion from project files
• /command completion from the running pi session
• saved session resume support
• model selection and thinking-level controls
• session compaction, abort, reload, and HTML export
• compact or verbose tool-result rendering
• optional markdown-mode rendering for session buffers

Current workflow

The current package centers on a scope-bound session buffer:

1. open a session for the current buffer
2. compose a prompt in Emacs
3. send it to the hidden pi RPC process
4. watch streamed output in a side window
5. continue the conversation, resume old sessions, or switch models as needed

Configuration

Prompt composer buffers use org-mode by default. Set pi-ui-prompt-major-mode to another major mode, for example:

(setq pi-ui-prompt-major-mode #'markdown-mode)

Difference from dnouri/pi-coding-agent

This repo is closely related in spirit, but it is intentionally different in a few ways:

• smaller surface area — this repo aims to stay relatively small and easy to reason about
• buffer-oriented usage — it is biased toward lightweight usage from the current buffer instead of a richer chat workspace
• simpler window model — it uses a session side window plus an on-demand prompt composer, rather than a persistent chat-buffer/input-buffer pair as the core UX
• lighter UI ambitions — fewer integrated controls and less session-management UI than dnouri/pi-coding-agent

If you want the more established chat-first Emacs frontend, use dnouri/pi-coding-agent. If you want a smaller package that can evolve toward a tighter current-buffer workflow, this repo explores that direction.