Mindket

Cheap Agent

Newversion 1 · updated
Star310

Overview

Moves routine coding in Claude Code or Codex to a low-cost model that must prove its work with a test, so your main model's usage lasts longer.

Cheap Agent is an open-source project on GitHub in the Developer category, listed by Mindket. It is free. This is version 1, last updated on . Install it from GitHub as its README says. Get Mindket’s setup notes on your Mindket account, as a zip.

At a glance

You need
  • Claude Code or Codex
  • Node.js 20+
  • A funded API key (a few dollars is enough) from OpenRouter, DeepSeek, Command Code, OpenCode Go or OpenCode Zen, or a local model instead
  • On Windows, Git for Windows
Runs in
  • Claude Code: it installs itself for every project, then you allow it once in each project
  • Codex: it comes as a plugin from a marketplace, with three hooks you trust on first run
  • Windows: it also needs Git for Windows, which provides Git Bash
  • Providers: OpenRouter, DeepSeek, OpenCode Go, OpenCode Zen, Command Code or a local server
Start with
npx cheap-agent

Platform

Claude CodeCoding agent
CodexCoding agent
macOSOperating system
LinuxOperating system
WindowsOperating system
OpenCodeCoding agent

About

What it is

Cheap Agent splits coding work between two models in Claude Code or Codex. Your main model plans and decides. A much cheaper model does the typing in a private copy of your project and has to show a passing test before you see any of it. The aim is to make your Claude or ChatGPT plan last longer.

Who it suits

It suits developers who run into plan limits because a top model spends its usage on routine typing.

IndustryWho uses itHow they use itExample
Software developmentDevelopers on a Claude or ChatGPT planLet a cheap model do the typing while the main model plansOne feature built with 57k Opus tokens instead of 154k
EducationStudents who cannot afford a $100 planStretch a plan they already pay forThe project aims at $50 to 70 of value from a $20 plan
Small teamsTeams watching their AI spendPay a few dollars of API usage for the typingA few dollars lasting hundreds of tasks

Anyone who uses neither Claude Code nor Codex can skip it.

Where to use it

It adds itself to Claude Code or Codex and runs on macOS, Linux and Windows.

  • Claude Code: it installs itself for every project, then you allow it once in each project
  • Codex: it comes as a plugin from a marketplace, with three hooks you trust on first run
  • Windows: it also needs Git for Windows, which provides Git Bash
  • Providers: OpenRouter, DeepSeek, OpenCode Go, OpenCode Zen, Command Code or a local server
Use caseWhoWhat they get back
Build a featureDevelopersA tested change from a cheap model that worked in a separate copy of the project
Save plan usagePeople who hit usage limitsFewer Opus tokens spent on the same feature
Choose the modelsCost-conscious usersBalanced, cheapest or strongest presets, or your own provider and model

How to set it up

You need Claude Code or Codex, Node.js and an API key before you start.

  • Claude Code or Codex
  • Node.js 20+
  • A funded API key (a few dollars is enough) from OpenRouter, DeepSeek, Command Code, OpenCode Go or OpenCode Zen, or a local model instead
  • On Windows, Git for Windows
  1. Run the login command and pick a provider. Choose OpenRouter if you are unsure.
  2. Press Enter for the recommended models and paste your API key.
  3. It tests your key, then registers itself in Claude Code across all your projects.
  4. In each project, run the allow command once.
npx cheap-agent

Then, once in each project, inside Claude Code:

/cheapag:allow

For Codex, log in with the same command. If you do not use Claude, decline when it offers to install itself into Claude Code. Then add the plugin:

codex plugin marketplace add muaaz9329/cheap-ag
codex plugin add cheapag@cheap-agent

Launch codex inside your project. When it asks about the 3 cheapag hooks, trust them. Then, once in each project:

$cheapag:allow

If you prefer, paste a setup prompt from the README into Claude Code. It asks for your provider and models, then hands you one command to run.

How to use it

Ask Claude or Codex for changes as you always do. The main model plans, the cheap one writes the code in a separate copy of the project, and a test has to pass before the result reaches you.

ToolWhat it does
npx cheap-agentLogs you in, checks your key and adds Cheap Agent to Claude Code
/cheapag:allowRun once in each project before you use it in Claude Code
$cheapag:allowThe same one-time step in Codex
claude mcp listShows cheapag when the Claude Code install worked

Example

You ask Claude Code to add a feature as usual. In the README's side-by-side test of one feature, Claude Code alone spent 154k Opus tokens over 5 turns. With Cheap Agent the same job took 57k Opus tokens over 2 turns, a 63% cut, plus about half a cent on the cheap model. Both versions passed their tests.

For newcomers and experts

If you pay for Claude or ChatGPT, your plan has a usage limit, and a big model spends part of it on routine typing. Cheap Agent hands that typing to a far cheaper model. Install it for Claude Code first, then ask for one small change.

Underneath, your main model plans and decides while a cheap model, run through the OpenCode engine, implements in its own copy of the project. It is listed as tested on SWE-bench Verified. The author says it will not make Opus ten times cheaper. The cheap model also cannot read dependency folders such as node_modules in its copy, though your tests still run normally.

  1. Your main model plans and decides
  2. A cheap model gets the task in its own project copy
  3. It writes the code and proves it with a test
  4. The result reaches you once the test passes

Good to know

  • Cost: a few dollars on an API key lasts hundreds of tasks
  • Provider notes: a Command Code key needs a plan that includes API access, and for the DeepSeek models OpenCode Go wants its console Regions set to Global
  • Windows start-up: the first job takes roughly 2 minutes while the OpenCode engine installs itself
  • Windows crashes: a crashed engine's leftover copy may keep running as opencode.exe until you close it or restart
  • Numbers: the token counts come from the author's own side-by-side runs
  • License: MIT
Opus tokens in the README's two tests
Test 1, Claude Code alone154 thousand tokens
Test 1, with Cheap Agent57 thousand tokens
Test 2, Claude Code alone233 thousand tokens
Test 2, with Cheap Agent132 thousand tokens

README and files

README

Cheap Agent

Moves routine coding in Claude Code or Codex to a low-cost model that must prove its work with a test, so your main model's usage lasts longer.

Made by muaaz9329 on GitHub. These notes are Mindket's, written from its README; the project's own files and license are in its repository.

About

What it is: Cheap Agent splits coding work between two models in Claude Code or Codex. Your main model plans and decides. A much cheaper model does the typing in a private copy of your project and has to show a passing test before you see any of it. The aim is to make your Claude or ChatGPT plan last longer.

Who it suits: It suits developers who run into plan limits because a top model spends its usage on routine typing.

Industry | Who uses it | How they use it | Example
Software development | Developers on a Claude or ChatGPT plan | Let a cheap model do the typing while the main model plans | One feature built with 57k Opus tokens instead of 154k
Education | Students who cannot afford a $100 plan | Stretch a plan they already pay for | The project aims at $50 to 70 of value from a $20 plan
Small teams | Teams watching their AI spend | Pay a few dollars of API usage for the typing | A few dollars lasting hundreds of tasks

Anyone who uses neither Claude Code nor Codex can skip it.

Where to use it: It adds itself to Claude Code or Codex and runs on macOS, Linux and Windows.

  • Claude Code: it installs itself for every project, then you allow it once in each project
  • Codex: it comes as a plugin from a marketplace, with three hooks you trust on first run
  • Windows: it also needs Git for Windows, which provides Git Bash
  • Providers: OpenRouter, DeepSeek, OpenCode Go, OpenCode Zen, Command Code or a local server
Use case | Who | What they get back
Build a feature | Developers | A tested change from a cheap model that worked in a separate copy of the project
Save plan usage | People who hit usage limits | Fewer Opus tokens spent on the same feature
Choose the models | Cost-conscious users | Balanced, cheapest or strongest presets, or your own provider and model

How to set it up: You need Claude Code or Codex, Node.js and an API key before you start.

  • Claude Code or Codex
  • Node.js 20+
  • A funded API key (a few dollars is enough) from OpenRouter, DeepSeek, Command Code, OpenCode Go or OpenCode Zen, or a local model instead
  • On Windows, Git for Windows
  1. Run the login command and pick a provider. Choose OpenRouter if you are unsure.
  2. Press Enter for the recommended models and paste your API key.
  3. It tests your key, then registers itself in Claude Code across all your projects.
  4. In each project, run the allow command once.
npx cheap-agent

Then, once in each project, inside Claude Code:

/cheapag:allow

For Codex, log in with the same command. If you do not use Claude, decline when it offers to install itself into Claude Code. Then add the plugin:

codex plugin marketplace add muaaz9329/cheap-ag
codex plugin add cheapag@cheap-agent

Launch codex inside your project. When it asks about the 3 cheapag hooks, trust them. Then, once in each project:

$cheapag:allow

If you prefer, paste a setup prompt from the README into Claude Code. It asks for your provider and models, then hands you one command to run.

How to use it: Ask Claude or Codex for changes as you always do. The main model plans, the cheap one writes the code in a separate copy of the project, and a test has to pass before the result reaches you.

Tool | What it does
npx cheap-agent | Logs you in, checks your key and adds Cheap Agent to Claude Code
/cheapag:allow | Run once in each project before you use it in Claude Code
$cheapag:allow | The same one-time step in Codex
claude mcp list | Shows cheapag when the Claude Code install worked

Example: You ask Claude Code to add a feature as usual. In the README's side-by-side test of one feature, Claude Code alone spent 154k Opus tokens over 5 turns. With Cheap Agent the same job took 57k Opus tokens over 2 turns, a 63% cut, plus about half a cent on the cheap model. Both versions passed their tests.

For newcomers and experts: If you pay for Claude or ChatGPT, your plan has a usage limit, and a big model spends part of it on routine typing. Cheap Agent hands that typing to a far cheaper model. Install it for Claude Code first, then ask for one small change.

Underneath, your main model plans and decides while a cheap model, run through the OpenCode engine, implements in its own copy of the project. It is listed as tested on SWE-bench Verified. The author says it will not make Opus ten times cheaper. The cheap model also cannot read dependency folders such as node_modules in its copy, though your tests still run normally.

Your main model plans and decides
A cheap model gets the task in its own project copy
It writes the code and proves it with a test
The result reaches you once the test passes

Good to know:

  • Cost: a few dollars on an API key lasts hundreds of tasks
  • Provider notes: a Command Code key needs a plan that includes API access, and for the DeepSeek models OpenCode Go wants its console Regions set to Global
  • Windows start-up: the first job takes roughly 2 minutes while the OpenCode engine installs itself
  • Windows crashes: a crashed engine's leftover copy may keep running as opencode.exe until you close it or restart
  • Numbers: the token counts come from the author's own side-by-side runs
  • License: MIT
Title: Opus tokens in the README's two tests
Test 1, Claude Code alone: 154 thousand tokens
Test 1, with Cheap Agent: 57 thousand tokens
Test 2, Claude Code alone: 233 thousand tokens
Test 2, with Cheap Agent: 132 thousand tokens

On GitHub, 7 November 2026

  • Stars: 310
  • Forks: 29
  • Open issues and pull requests: 2
  • Language: TypeScript
  • License: MIT
  • Topics: agentic-coding, ai-agents, anthropic, cheap-llm, claude, claude-code, claude-code-hooks, claude-code-plugin, claude-code-skill, cost-optimization, deepseek, developer-tools

Links

Get the code

git clone https://github.com/muaaz9329/cheap-ag.git

Then follow the setup in its README. Nothing in this kit runs by itself.

Files
README.md  6 KB

1 file, 6 KB, downloaded as a zip.

What’s new

  • Version 1 · 2026-11-07

Compare with similar

Listed GitHub projects in the same category that share the most topics with this one.

ProductStarsActivityLanguageLicenseLast change
Cheap Agent This page310ActiveTypeScriptMIT
Context ModeShares claude-code-hooks · claude-code-skill · claude33,650—TypeScript——
Claude-Mem by thedotmackShares claude-code-plugin · anthropic · claude116,866ActiveTypeScriptApache-2.0
PonytailShares claude-code-plugin · developer-tools · claude186,952ActiveJavaScriptMIT
ECC by affaan-mShares anthropic · developer-tools · claude299,210ActiveJavaScriptMIT
Recursive Decomposition SkillShares claude-code-skill · claude-code-plugin · ai-agents54——MIT—
Numbers as Mindket last read them, from 2026-11-07 to 2026-11-14. Activity is read from each project’s last change on GitHub: Active within 30 days, Quiet within 180, Stale after that.

Reviews

No ratings yet

Write a review

No reviews yet.

More from Mindket

See all from Mindket

anaksor-ocr

Sokheng · Kit

Reads Khmer and English text from documents, receipts and photos of signs, using a detector for text lines and a recogniser for each line. Suits developers who…

New

Qwen2.5-7B-Instruct as GGUF files in eight quantization sizes, to chat with locally through llama.cpp. For people who want a capable 7.6B model on their own co…

New

ACE-Step 1.5 XL SFT

ACE-Step · Kit

Generates music from text with a 4B diffusion decoder, 50 steps and guidance control, aiming for higher audio quality than the base. For creators with a GPU of…

New

Similar projects

See all in Developer

ECC by affaan-m

affaan-m · Skill

Affaan Mustafa’s plugin of agents, skills, commands and hooks that gives Claude Code, Codex and other coding agents a routine of planning, test-first work, fre…

New299k

Ponytail

dietrichgebert · Skill

Guides AI coding agents to write minimal code by prioritizing existing repository components and standard libraries. It adds auditing, code review, and trackin…

New187k

Claude-Mem by thedotmack

thedotmack · Kit

Alex Newman’s memory plugin for Claude Code and other coding agents. It records what the agent does in each session, turns that into short searchable notes and…

New117k