@btseytlin/pi-duck-mode

pi extension to spend 14x less tokens and achieve 798x more TPS

Install • What duck does • How it works • Benchmark • Commands • Troubleshooting

Installation

pi install npm:@btseytlin/pi-duck-mode

Or from git:

pi install git:github.com/btseytlin/pi-duck-mode

Type /duck to enter duck mode. In duck mode, no LLM calls are made and every reply is "Great, go ahead and make it!". Type your prompt, then write the code.

14x less tokens, 798x more TPS, <2ms overhead. Zero hallucinations. It also prevents developer skill atrophy.

Not affiliated. This project is not affiliated with, endorsed by or connected to DuckDB or any other project named duck.

What duck does

How Savings Work

duck cuts 100% of the LLM output and input. You still have to write the code.

How It Works

  Without duck:                                With duck:

  You  --prompt-->  pi  -->  LLM               You  --prompt-->  pi  -->  duck
   ^                          |                 ^                          |
   |   2503 ms, 99 tokens     |                 |   1.3 ms, 7 tokens       |
   +--------------------------+                 +--------------------------+

Four strategies applied to every prompt:

1. Smart Filtering - Removes noise (the answer).
2. Grouping - Aggregates similar replies (all of them) into one.
3. Truncation - Keeps relevant context, cuts the reply to 7 tokens.
4. Deduplication - Collapses repeated replies. There is only ever one.

Unlike other tools, duck mode doesn't break prompt caching. There is nothing to cache.

Commands

/duck                           # Enter duck mode, or leave it
/quit                           # Exit pi
/exit                           # Exit pi, Claude code style

Benchmark

Prompt: "Write a binary search function in Python. Return the index of the target, or -1."