convez.ai
AgenteAutomaçãoIntermediário
JosefStack

Codie - Agent CLI de Codificação com IA

JosefStack/codie

Agent CLI que escreve, testa e depura código sem sair do terminal

Ver no GitHub
InicialPluginScore · de 0 a 100

Agent CLI alimentado por IA que lê, escreve e depura código diretamente no terminal. Funciona como Claude Code, mas executável localmente, com acesso a shell, busca web, linting e testes automáticos.

Instalação

$npx convez-ai JosefStack/codie

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
1
Forks
0
Acompanhando
1
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Gerar APIs REST completas com testes integrados
  • Debugar e refatorar código existente automaticamente
  • Automatizar tarefas repetitivas de desenvolvimento

Veredito

Codie é um agent produtivo para desenvolvimento local, com bom controle de execução (review/auto/plan) e integração com ferramentas reais (ripgrep, shell, web search). Ideal para quem quer assistência de IA no workflow CLI sem depender de plataformas externas.

README

A tradução para o português deste README ainda está na fila.

Codie

An AI-powered CLI coding agent. Reads, writes, and debugs your code.

Codie Banner


What is Codie?

Codie runs in your terminal. You describe a task, it reads your files, writes code, runs commands, searches the web, and verifies its work — all without leaving your terminal.

→ build a REST API with FastAPI, add tests, and make sure it runs
  ⚙ Tool call: write_file
  ⚙ Tool call: write_file
  ⚙ Tool call: run_command
  ⚙ Tool call: run_debug
Codie: Done. Server running on port 8000, all tests passing.

Features

  • File tools — read, write, edit, delete files in your project
  • Shell access — run commands with mode-aware confirmation
  • Code search — ripgrep-powered search across your codebase
  • Web tools — search the web and read documentation pages
  • Debug loop — automatically runs lint and tests after writing code
  • Project memory — remembers your stack, conventions, and key files across sessions
  • Token tracking — see exactly how much each session costs with /cost
  • Model switching — switch between models mid-session with /model

Requirements

  • Python 3.10+
  • ripgrep — install before using Codie:
    • Windows: winget install BurntSushi.ripgrep.MSVC
    • Mac: brew install ripgrep
    • Linux: apt install ripgrep
  • A Groq API key (free)
  • A Tavily API key (free, 1000 searches/month)
  • A Jina API key (free, 500 RPM)

Installation

pip install codie-cli

First Run

On first run, Codie walks you through setup:

First run setup

To update your keys or change model at any time:

codie --configure

Usage

codie                  # start in default mode
codie --mode auto      # start in auto mode
codie --mode plan      # start in plan mode
codie --version        # show version
codie --configure      # update API keys and settings

Modes

ModeBehavior
reviewConfirms each file edit and mutating shell command
autoRuns without asking (still confirms destructive commands)
planShows full plan, user approves once, then executes

Slash Commands

Slash commands

CommandDescription
/helpShow available commands
/mode <mode>Change mode mid-session
/modelShow current model and available models
/model <model>Switch model mid-session
/costShow token usage and cost for this session
/remember <thing>Save something to project memory
/clearClear the screen
/exitExit Codie

Model Switching

Current/Available Models Model switching

Switch models mid-session without restarting:


Mode Switching

ModeBehavior
reviewConfirms each file edit and mutating shell command
autoRuns without asking (still confirms destructive commands)
planShows full plan first, user approves once, then executes all

Mode switching

Project Memory

Codie can maintain a .codie/AGENT.md file in your project — your stack, conventions, important files, and notes injected into every session.

codie
# Setup project memory? [y/n]: y
# Project memory created at .codie/AGENT.md

Add .codie/AGENT.md to git to share context with your team.

Use /remember to add notes mid-session:

→ /remember always use TypeScript strict mode

Token Tracking

Cost tracking


Configuration

Config is stored at ~/.codie/config.toml. Run codie --configure to update.

You can also use environment variables:

export GROQ_API_KEY=...
export TAVILY_API_KEY=...
export JINA_API_KEY=...
export CODIE_MODEL=openai/gpt-oss-120b

License

MIT — built by Joseph Johnson

Ver o original no GitHub

Bom para

Relacionados