A plugin-based harness that gives Claude Code, Codex and other coding agents shared agents, skills, hooks, memory and security scanning.
rtk-ai/rtk
CLI proxy that filters and compresses shell command output before it reaches an AI coding agent's context, cutting token use.
- Estrelas
- 83,930
- Forks
- 5,334
- Issues abertas
- 1,543
- Último commit
- há 10 horas
O que é?
RTK sits between an AI coding agent and the shell. When the agent runs commands such as git status, git diff, grep, ls, cargo test or pytest, RTK rewrites the output into a shorter form: failures only for test runs, grouped matches for search, file counts for directory listings, and trimmed diffs. The stated goal is a 60-90% reduction in tokens for these common developer commands, which lowers context cost for the agent.
It ships as a single Rust binary and covers 100+ commands across Git, Cargo, npm, Python, Go, Docker, Kubernetes, AWS and Terraform. Integration is handled by an init command that installs hooks for supported agents, including Claude Code, Cursor, Gemini CLI, Codex, Windsurf, Cline and Copilot, so the agent receives compact output without calling rtk explicitly. The project also offers rtk gain for reporting savings and an optional rtk-proxy daemon for compressing provider API traffic.
The project is dual-licensed: the rtk CLI is Apache-2.0, while the compression engine and proxy are under BUSL-1.1, with a separate Apache-2.0-only edition available.
Recursos principais
- 01Rewrites common shell commands such as git, grep, ls and test runners into compact output.
- 02Ships as a single Rust binary with no runtime dependencies.
- 03Installs agent hooks for Claude Code, Cursor, Gemini CLI, Codex, Copilot and many others.
- 04Reports token savings with rtk gain, including history and JSON export.
- 05Offers rtk bypass to run a single command unfiltered while still tracking it.
Casos de uso
- Reducing token cost when running an AI coding agent on large repositories with verbose test or build output.
- Keeping an agent's context focused by collapsing passing tests and noisy git diffs.
- Measuring which commands would benefit from filtering using rtk discover and rtk gain.
Tópicos
Perguntas frequentes
- O que é rtk?
- CLI proxy that filters and compresses shell command output before it reaches an AI coding agent's context, cutting token use. RTK sits between an AI coding agent and the shell. When the agent runs commands such as git status, git diff, grep, ls, cargo test or pytest, RTK rewrites the output into a shorter form: failures only for test runs, grouped matches for search, file counts for directory listings, and trimmed diffs. The stated goal is a 60-90% reduction in tokens for these common developer commands, which lowers context cost for the agent.
- rtk é gratuito? Qual é a licença?
- rtk é open source sob a licença Apache-2.0. Confira os termos da licença antes de usá-lo em produtos comerciais.
- rtk ainda é mantido?
- Manutenção ativa. O último commit de rtk foi em 19 de out. de 2026.
- Para que posso usar rtk?
- Reducing token cost when running an AI coding agent on large repositories with verbose test or build output. Keeping an agent's context focused by collapsing passing tests and noisy git diffs. Measuring which commands would benefit from filtering using rtk discover and rtk gain.
- Onde consigo rtk?
- O código-fonte e as instruções de instalação estão no GitHub: https://github.com/rtk-ai/rtk
Resumo gerado por IA a partir do README do projeto. As estatísticas do GitHub são atualizadas diariamente.
Projetos relacionados
Minimal Rust-based statusline HUD for Claude Code showing model, Git state, workspace, and context window usage.
A macOS menu bar app that shows live Claude Code status: an animated icon while working, a yellow dot when it needs your permission.
A Claude Code plugin that shows short replies by default, with phrases that expand in place when clicked.
A CLI that scans your codebase and generates compact AI context files so coding assistants spend fewer tokens exploring the project.
A VS Code extension that lets Claude Code, OpenAI Codex and other coding agents answer, review and debate in one conversation.