pi-context-whisperer 🦜

Smart gradual auto-compaction for Pi — warns you and auto-compacts before context limits hit, preserving your flow.

How it works

Context Whisperer monitors your context window usage after every turn and takes action before you hit limits:

Context: 45% → 🦜 ███████░░░ 58k/128k ✓ Healthy
Context: 72% → 🦜 █████████░░ 92k/128k ⚡ Warning    (notifies you)
Context: 82% → 🦜 ██████████░ 105k/128k ⚠ COMPACT NOW (auto-compacts)
Context: 35% → 🦜 █████░░░░░ 45k/128k ✓ Healthy     (freed up!)

Features

• Context health bar — live in footer: ████░░░░ with percentage
• Warning at customizable threshold — nudges you when context gets tight
• Auto-compaction — automatically compacts at critical threshold so you don't crash
• Prevents wasted calls — no more "context limit exceeded" mid-task
• Compaction counter — see how many times it's saved you this session
• Fully configurable — set your own warn/auto thresholds

Install

# npm
pi install npm:pi-context-whisperer

# GitHub
pi install git:github.com/Jaraxxxx/pi-context-whisperer

Commands

Command	Description
/whisper-enable	Turn on auto-compaction
/whisper-disable	Turn off (manual /compact only)
/whisper-stats	Show current stats and thresholds
/whisper-config 60 80	Set warn at 60%, auto at 80%
Ctrl+Shift+C	Force compact now

LLM Tools

The context_health tool lets the agent check context status before making large requests.

Default thresholds

Threshold	Default	Behavior
Warn	70%	Notification only
Auto-compact	80%	Triggers compaction automatically

Change with: /whisper-config <warnPct> <autoPct>

Requirements

• Pi coding agent
• Auto-compaction enabled in settings (default: on)