TokenEconomizer: Cut Coding Agent Costs With Experience-Driven Control

TokenEconomizer: Cut Coding Agent Costs With Experience-Driven Control

GitHub Open Source
microsoft/TokenEconomizer

Cut coding agent costs by up to 30.9% across repeated repository tasks without altering the agent.

Repository snapshot

  • Repository: microsoft/TokenEconomizer
  • Language: Python
  • License: MIT
  • Stars: 7 666
  • Forks: 525

Daily open-source picks: GitHub Open Source on Telegram

What it is

TokenEconomizer is an open-source plugin designed to reduce token costs for AI coding agents without modifying the underlying model or weights. It operates as a black-box optimizer that analyzes past execution logs within a specific repository, learning what works and what wastes resources across multiple sessions.

Why it matters

Coding agents repeatedly spend tokens on redundant instructions, unused tool definitions, and forgotten repository quirks. By capturing these patterns locally, TokenEconomizer eliminates unnecessary overhead and cuts mean execution costs significantly across repeated tasks.

Key features

  • Inject micro-skills into the session context once verified across multiple runs.
  • Prune unused tool definitions based on configurable prompt and session thresholds.
  • Detect recurring failure loops and provide constructive redirects.
  • Store all learned repository state locally on your machine.
  • Support major assistants including Claude Code, Codex, and GitHub Copilot CLI.

Quick start

Clone the repository to get started with supported agents:

git clone https://github.com/microsoft/TokenEconomizer.git

Who should try it

Developers who frequently rely on AI coding assistants and want to minimize recurring API expenses without changing their core workflow.

Verdict

A practical, transparent way to trim wasted tokens for heavy agent users.

Continue exploring

Get the next pick in Telegram: Follow GitHub Open Source

Curated from the repository README and metadata. No hands-on test is claimed; evaluate it in a sandbox before production use.

Report Page