TokenTrim — Output Compressor for Coding Agents

Cut coding-agent tool-call output tokens by ~30% and save on API bills without hurting agent performance

Problem

Heavy Codex/Astra users generate huge token volumes from tool-call outputs—one person burned $700/day on API—and subscription quotas run out fast. Cost pressure is real and recurring.

Solution

A local CLI or proxy layer that intercepts and smartly compresses agent tool outputs (logs, file contents, diffs), stripping redundant lines and repeated context while keeping only what the next reasoning step needs, with verification that downstream actions still work.

Users

Developers who use coding agents daily, small teams paying per API call, and AI product teams.

MVP

Rule-plus-model compression for 3–5 common output types (file reads, terminal logs, git diffs, directory listings); local CLI plugging into Claude Code/Codex; before/after token and dollar comparison; runs fully locally with no code upload.

Revenue

Free CLI plus paid Pro (more tool types, team usage dashboard), or a subscription priced as a percentage of savings ($15–30/mo).

Why now

Coding-agent usage is exploding, API bills are a top pain point, and compression tooling is still scarce—an indie dev can validate fast and win users with real savings numbers.