Skip to content
ToolOpen-source agent

TokenEconomizer

Learns from past agent sessions to cut coding-agent token costs

Pricing
Open source
Platforms
CLI
GitHub stars
7.2K
Last commit
today
Language
Python
License
Apache-2.0

About TokenEconomizer

TokenEconomizer is an open-source AI tool for developer tools. It is available on CLI.

This wraps an existing coding agent rather than replacing it. It watches sessions in a repository and applies what it learns to later ones, through three mechanisms.

Micro-skills record repeated repo facts, such as the right test command, once three sessions confirm them. Tool pruning leaves unused tool definitions out of future requests through the agent's own settings, and one command restores them. A loop guard asks a judge model whether a turn is stuck and, if confident, sends one redirect note.

The authors report 18.4 to 30.9 percent lower mean cost on SWE-Bench tasks across five agent and model pairs. That figure comes from the project's own measurement. It is a Microsoft and UC Berkeley research project, Apache-2.0 licensed.

What it does

Micro-skills
Stores repo facts confirmed across three sessions.
Tool pruning
Drops tools that go unused over several sessions.
Loop guard
Judge model can redirect a stuck turn once.
Black-box control
Works without changing the agent itself.
Reversible
One command restores pruned tools.

Good for

  • Lower token spend on a frequently used repo
  • Stop agents repeating the same failed command
  • Reduce tool-definition overhead

Pricing

Open source

Apache-2.0

TokenEconomizer FAQ

Does it modify the agent?

No, it works without changing the agent itself.

How much does it save?

The authors report 18.4 to 30.9 percent lower mean cost on SWE-Bench tasks.

Will it stop a stuck turn?

It redirects at most once, then tells you how to stop it.

Is TokenEconomizer free?

Yes. TokenEconomizer is open source, so you can use and self-host it for free.

What are the best alternatives to TokenEconomizer?

The closest alternatives to TokenEconomizer are ECC, firecrawl, ponytail, graphify and orca.

Alternatives to TokenEconomizer

  1. Skills, hooks, memory and security layer for coding agent harnesses

    • Freemium
    • Open-source agent
    • 305K GitHub stars
    • AI models and LLM tooling
  2. Web search, scraping and page interaction API for AI agents and developers

    • Freemium
    • Agent framework
    • 202K GitHub stars
    • Web search and scraping
  3. Ruleset that makes AI coding agents write less code that still works

    • Open source
    • Open-source agent
    • 192K GitHub stars
    • AI models and LLM tooling
  4. Code knowledge graph that gives coding assistants shared, source-linked context

    • Freemium
    • Open-source agent
    • 149K GitHub stars
    • Knowledge and memory
  5. Desktop environment for running several coding agents in parallel worktrees

    • Open source
    • Open-source agent
    • 127K GitHub stars
    • Coding
  6. Memory for coding agents that saves decisions and fixes, with optional hosted sync

    • Freemium
    • Open-source agent
    • 118K GitHub stars
    • Knowledge and memory