basic

Compact: Basic

Compact a long agent transcript and print the before/after token counts.

Setup

TypeScript:

npm install

Python:

pip install -r requirements.txt

Run

MORPH_API_KEY=your-key npx tsx compact.ts
MORPH_API_KEY=your-key python compact.py

What it does

1. Sends a ~560 token transcript to POST /v1/compact with query, compression_ratio: 0.4, and preserve_recent: 0
2. Prints the compacted output, where dropped runs appear as (filtered N lines) markers
3. Prints usage — input tokens, output tokens, and the achieved compression ratio

Real output from a run:

input tokens:       559
output tokens:      259
compression ratio:  0.463
processing time:    67ms

The transcript wanders across five topics; the query names one of them. Everything about CI config, staging seed data, and the unrelated CSS bug gets dropped. The idempotency-key fix, the duplicate-charge SQL, and the alerting plan survive — verbatim, character for character.

preserve_recent: 0 is used here so the whole transcript is eligible for compaction. In production set it to 3 or higher so the latest turns pass through untouched.