开源 · MIT · Rust 50.3k stars

你的 AI 代理正淹没在
CLI 噪音中。 解决它。

rtk 在命令输出进入上下文窗口之前对其进行压缩。 更好的推理。更长的会话。更低的成本。

89% 平均去除噪音
3x 更长会话
30+ 命令数
Why RTK? The numbers.

89% noise reduction measured across 2,900+ real-world dev commands: cargo test 91.8%, git status 80.8%, find 78.3%, grep 49.5%. Free, open source (MIT, Rust).

安装 rtk GitHub 上加星
Viking warrior smashing token noise with his axe
Terminal
$ rtk gain

📊 RTK Token Savings
════════════════════════════════════════

Total commands:    2,927
Input tokens:      11.6M
Output tokens:     1.4M
Tokens saved:      10.3M (89.2%)

By Command:
────────────────────────────────────────
Command               Count      Saved     Avg%
rtk find                324       6.8M    78.3%
rtk git status          215       1.4M    80.8%
rtk grep                227     786.7K    49.5%
rtk cargo test           16      50.1K    91.8%

$
Works with
  • Claude Code
  • Cursor
  • Aider
  • Gemini CLI
  • OpenAI Codex
  • Cline
  • Windsurf
  • GitHub Copilot
// context forge

一个锻造厂。 三个工具。

RTK、ICM 和 Vox 共享一种理念：开源、Rust、零遥测、本地优先。

RTK 压缩

命令、文件读取、测试——在到达模型上下文之前压缩 60-90%。零配置。

89% 噪音消除
ICM 记忆

跨会话的持久记忆。你的代理从上次停下的地方继续——决策、错误、上下文。

无限上下文记忆
Vox 说话

为 AI 代理提供语音输出。三个 TTS 后端，四种 Claude Code 集成模式。

3 个 TTS 后端
// 01 — 问题所在
Viking bracing against a tidal wave of tokens

当今 AI 编程的问题

代理执行的每个命令都会用噪音污染上下文窗口。这会产生以下代价。

上下文污染

你的 200K 上下文窗口不是无限的。当 cargo test 输出 5000 个模板 token 时，就少了 5000 个用于推理实际代码的 token。

context_quality: degraded ▼

会话太短

上下文溢出，代理重启，你失去了线索。在固定费率计划上，你会比应该的速度快 40% 达到速率限制。

session_remaining: 32% ▼

成本爆炸

在按 token 付费的设置中（API、Gemini CLI、Aider），70% 的账单是 LLM 不需要的噪音。10 人团队每月浪费约 $1,750。

token_waste: $1,750/mo ▲
Viking compressing tokens into a glowing crystal

查看差异

真实输出，真实节省。实际命令的并排对比。

cargo test ~4,823 tokens 
warning: unused variable: `start`
   --> src/init.rs:561:17
    |
561 |             let start = i;
    |                 ^^^^^ help: prefix it with an underscore

warning: `rtk` (bin "rtk" test) generated 17 warnings
    Finished `test` profile target(s) in 0.20s
     Running unittests src/main.rs

running 262 tests
test cargo_cmd::tests::test_filter_cargo_build_success ... ok
test cargo_cmd::tests::test_filter_cargo_clippy_clean ... ok
test cargo_cmd::tests::test_filter_cargo_test_all_pass ... ok
test cargo_cmd::tests::test_filter_cargo_test_failures ... ok
test cc_economics::tests::test_compute_dual_metrics ... ok
test git::tests::test_compact_diff ... ok
test git::tests::test_filter_branch_output ... ok
test git::tests::test_filter_log_output ... ok
test git::tests::test_filter_status_with_args ... ok
... 253 more tests, all passing

test result: ok. 262 passed; 0 failed; 0 ignored; 0 measured
rtk cargo test ~11 tokens -99% 
✓ cargo test: 262 passed (1 suite, 0.08s)
pytest -v ~756 tokens 
===== test session starts ======
platform darwin -- Python 3.14.3, pytest-9.0.2
collected 33 items

test_utils.py::TestStringUtils::test_strip PASSED    [  3%]
test_utils.py::TestStringUtils::test_upper PASSED    [  6%]
test_utils.py::TestStringUtils::test_split PASSED    [  9%]
test_utils.py::TestMathUtils::test_sqrt PASSED       [ 27%]
test_utils.py::TestMathUtils::test_ceil PASSED       [ 30%]
test_utils.py::TestJsonUtils::test_dumps PASSED      [ 51%]
test_utils.py::TestJsonUtils::test_loads PASSED      [ 54%]
test_utils.py::TestPathUtils::test_join PASSED       [ 63%]
test_utils.py::TestListOps::test_sort PASSED         [ 78%]
test_utils.py::TestListOps::test_flatten PASSED      [100%]
... 23 more tests, all passing

====== 33 passed in 0.05s ======
rtk test "pytest -v" ~24 tokens -96% 
33 passed in 0.02s
go test ./... -v ~592 tokens 
=== RUN   TestControllers
Running Suite: Controller Suite
Random Seed: 1771093287
Will run 1 of 1 specs
•
Ran 1 of 1 Specs in 6.037 seconds
SUCCESS! -- 1 Passed | 0 Failed | 0 Pending
--- PASS: TestControllers (6.04s)
PASS
ok  kubecraft.ai/.../controller  6.610s
=== RUN   TestNewAWSBedrockClient
--- PASS: TestNewAWSBedrockClient (0.00s)
=== RUN   TestNewAzureOpenAIClient
--- PASS: TestNewAzureOpenAIClient (0.00s)
--- PASS: TestAzureOpenAIGenerateCode_Success (0.00s)
--- PASS: TestNewMistralClient (0.00s)
--- PASS: TestGenerateCode_Success (0.00s)
PASS
ok  kubecraft.ai/.../llm  0.776s
rtk test "go test ./... -v" ~246 tokens -58% 
--- PASS: TestControllers (5.07s)
ok  kubecraft.ai/.../controller  5.847s
--- PASS: TestNewAWSBedrockClient (0.00s)
--- PASS: TestNewAzureOpenAIClient (0.00s)
--- PASS: TestAzureOpenAIGenerateCode_Success (0.00s)
--- PASS: TestNewMistralClient (0.00s)
--- PASS: TestGenerateCode_Success (0.00s)
ok  kubecraft.ai/.../llm  0.246s
git diff HEAD~1 ~21,500 tokens 
diff --git a/index.html b/index.html
index 1b7488b..0ebac4f 100644
--- a/index.html
+++ b/index.html
@@ -629,7 +629,7 @@
       width: 100%;
       border-collapse: collapse;
       font-size: 0.88rem;
-      min-width: 800px;
+      min-width: 1050px;
     }
@@ -1051,6 +1051,114 @@
+    /* === Share My Gain === */
+    .share-gain { background: var(--bg); }
+    .share-gain-card {
+      max-width: 600px;
+      margin: 0 auto;
... 855 insertions(+), 266 deletions(-)
rtk git diff HEAD~1 ~1,259 tokens -94% 
index.html | 1121 ++++++++++++------
 1 file changed, 855 ins(+), 266 del(-)

--- Changes ---

index.html
  @@ -629,7 +629,7 @@
  -      min-width: 800px;
  +      min-width: 1050px;
  @@ -1051,6 +1051,114 @@
  +    /* === Share My Gain === */
  +    .share-gain { background: var(--bg); }
  ... (truncated)
git status ~120 tokens 
On branch master
Your branch is up to date with 'origin/master'.

Changes not staged for commit:
  (use "git add <file>..." to update what will be committed)
  (use "git restore <file>..." to discard changes in working directory)
	modified:   index.html
	modified:   src/main.rs
	modified:   src/config.rs

Untracked files:
  (use "git add <file>..." to include in what will be committed)
	.fastembed_cache/
	tests/

no changes added to commit
rtk git status ~30 tokens -75% 
* master...origin/master
~ 3  index.html, src/main.rs, src/config.rs
? 2  .fastembed_cache/, tests/
git log --stat -10 ~1,430 tokens 
commit c84fa3c fix: add website URL (rtk-ai.app)
Author: patrick szymkowiak <52030887+pszymkowiak@...>
Date:   Thu Feb 12 12:28:37 2026 +0100

    - Cargo.toml: add homepage field
    - README.md: add website/github/install links

 .github/workflows/release.yml | 4 ++--
 Cargo.toml                    | 1 +
 Formula/rtk.rb                | 2 +-
 README.md                     | 8 ++++++++
 5 files changed, 14 insertions(+), 3 deletions(-)

commit a0d2184 feat(ci): automate Homebrew formula
... 8 more commits with full stats
rtk git log -n 10 ~194 tokens -86% 
c84fa3c fix: add website URL (rtk-ai.app) (#81)
a0d2184 feat(ci): automate Homebrew formula (#80)
55d010a fix: update stale repo URLs (#78)
9e764c4 chore(master): release 0.13.1
bd76b36 fix(ci): fix release workflow artifacts
0d388b3 chore(master): release 0.13.0
93364b5 feat(sqlite): custom db location
3dde699 chore(master): release 0.12.0
645a773 feat(cargo): cargo install filtering
22182e4 chore(master): release 0.11.0
cat src/main.rs ~10,176 tokens 
mod cargo_cmd;
mod cc_economics;
mod ccusage;
mod config;
mod container;
mod curl_cmd;
mod deps;
mod diff_cmd;
mod discover;
mod filter;
mod find_cmd;
mod gain;
mod gh_cmd;
mod git;
mod grep_cmd;
mod init;
mod json_cmd;
mod ls;
mod main.rs
... 1,295 lines total (enums, match arms, CLI parsing)
rtk read src/main.rs -l aggressive ~504 tokens -95% 
use anyhow::{Context, Result};
use clap::{Parser, Subcommand};
struct Cli {
    // ... implementation
enum Commands {
    // ... implementation
enum GitCommands {
    // ... implementation
enum CargoCommands {
    // ... implementation
fn main() -> Result<()> {
    // ... implementation
grep -rn "pub fn" src/ ~2,108 tokens 
src/cargo_cmd.rs:17:pub fn run(cmd: CargoCommand, ...) -> Result<()>
src/cargo_cmd.rs:551:pub fn run_passthrough(args: ...) -> Result<()>
src/cc_economics.rs:184:pub fn run(...)
src/config.rs:73:    pub fn load() -> Result<Self>
src/config.rs:85:    pub fn save(&self) -> Result<()>
src/config.rs:97:    pub fn create_default() -> Result<PathBuf>
src/filter.rs:57:    pub fn from_extension(ext: &str) -> Self
src/filter.rs:309:pub fn get_filter(level: FilterLevel) -> Box<dyn FilterStrategy>
... 112 matches across 47 files
rtk grep "pub fn" src/ ~940 tokens -55% 
112 in 47 files:

src/cargo_cmd.rs (3):
    17: pub fn run(cmd: CargoCommand, ...) -> Result<()>
   551: pub fn run_passthrough(args: ...) -> Result<()>

src/config.rs (4):
    73: pub fn load() -> Result<Self>
    85: pub fn save(&self) -> Result<()>
    97: pub fn create_default() -> Result<PathBuf>
   109: pub fn show_config() -> Result<()>

src/git.rs (2):
    22: pub fn run(cmd: GitCommand, ...) -> Result<()>
  1264: pub fn run_passthrough(args: ...) -> Result<()>
... +62
find . -name "*.rs" ~276 tokens 
./target/debug/build/serde_core-.../out/private.rs
./target/debug/build/libsqlite3-sys-.../out/bindgen.rs
./src/ls.rs
./src/local_llm.rs
./src/learn/detector.rs
./src/learn/report.rs
./src/discover/registry.rs
./src/discover/provider.rs
./src/cargo_cmd.rs
./src/ccusage.rs
./src/config.rs
./src/find_cmd.rs
./src/gain.rs
./src/git.rs
... 49 files total
rtk find "*.rs" . ~149 tokens -46% 
49 files in 4 dirs:

src/ cargo_cmd.rs cc_economics.rs
  ccusage.rs config.rs container.rs
  curl_cmd.rs deps.rs diff_cmd.rs
  filter.rs find_cmd.rs gain.rs
  gh_cmd.rs git.rs grep_cmd.rs
  init.rs json_cmd.rs ls.rs
  main.rs ...
src/discover/ mod.rs provider.rs
  registry.rs report.rs
src/learn/ detector.rs mod.rs report.rs
src/parser/ error.rs formatter.rs
ls -la src/ ~3,200 tokens 
total 928
drwxr-xr-x  41 patrick  staff   1312 .
drwxr-xr-x  25 patrick  staff    800 ..
-rw-r--r--   1 patrick  staff  16394 cargo_cmd.rs
-rw-r--r--   1 patrick  staff  27220 cc_economics.rs
-rw-r--r--   1 patrick  staff   9503 ccusage.rs
-rw-r--r--   1 patrick  staff   2884 config.rs
-rw-r--r--   1 patrick  staff  12886 container.rs
-rw-r--r--   1 patrick  staff   3406 curl_cmd.rs
-rw-r--r--   1 patrick  staff  12156 filter.rs
-rw-r--r--   1 patrick  staff  10648 gain.rs
... 20 more lines
rtk ls src/ ~640 tokens -80% 
discover/
parser/
cargo_cmd.rs  16.0K
cc_economics.rs  26.6K
ccusage.rs  9.3K
config.rs  2.8K
container.rs  12.6K
filter.rs  11.9K
find_cmd.rs  2.9K
gain.rs  10.4K
...

37 files, 2 dirs (37 .rs)
cat Cargo.toml ~368 tokens 
[package]
name = "rtk"
version = "0.13.1"
edition = "2021"
license = "MIT"
homepage = "https://www.rtk-ai.app"
repository = "https://github.com/rtk-ai/rtk"

[dependencies]
clap = { version = "4", features = ["derive"] }
anyhow = "1.0"
ignore = "0.4"
walkdir = "2"
regex = "1"
lazy_static = "1.4"
serde = { version = "1", features = ["derive"] }
serde_json = { version = "1" }
colored = "2"
dirs = "5"
rusqlite = { version = "0.31", features = ["bundled"] }
... + cargo-deb, cargo-generate-rpm metadata
rtk deps ~55 tokens -85% 
Rust (Cargo.toml):
  Dependencies (15):
    clap (4)
    anyhow (1.0)
    ignore (0.4)
    walkdir (2)
    regex (1)
    lazy_static (1.4)
    serde (1)
    serde_json (1)
    colored (2)
    dirs (5)
    ... +5 more
Viking king on throne with 88.9% holographic display

真实节省

满意的开发者通过 rtk gain 的真实反馈。

开发者反馈

日常使用几周后：15,720 个命令，节省 1.38 亿 token。运行 rtk gain 查看你的。

88.9% efficiency
rtk gain dashboard showing 88.9% efficiency

详细明细

按命令的每日、每周和每月统计。跟踪你的节省。

Per-command analytics
rtk gain daily and weekly breakdown
Viking squad marching through datacenter corridor

没有 AI 工具提供无限使用。每个 token 都很重要。

即使每月 $200，每个工具都有上限。RTK 压缩 CLI 噪音，让你的额度更持久。

一个典型的 2 小时 AI 代理编程会话：

~60 CLI 命令执行数
~210K CLI 噪音 token
~23K 使用 RTK（89% 减少）

没有 RTK，仅 CLI 输出就可能溢出 200K 上下文窗口。基于 2,900+ 个真实命令测量的平均 3,500 tokens/命令。

Claude Code Terminal
价格$20 — $200/mo
限制约 45 条/5小时（Pro），Max 5-20倍
上下文200K tokens
会话长约 3 倍
即使 Max $200/月也有每周上限。RTK 平均压缩 89% CLI 输出，配额延长约 3 倍。
Cursor IDE
价格$20 — $200/mo
限制$20 额度/月（Pro），约 225 Claude 请求
上下文最高 200K（Max 模式）
额度持久约 2 倍
即使 Ultra $200/月也有上限。RTK 压缩 CLI 输出，让每个请求更干净。
OpenAI Codex Agent
价格$20/月（Plus）— $200/月（Pro）
限制每 5 小时 30-1,500 条
上下文192K tokens
每个上限更多迭代
包含在 ChatGPT 计划中。Pro $200/月上限 1,500 条/5小时。RTK 压缩命令输出，获得更多迭代。
Windsurf IDE
价格$15 — $60/mo
限制500 额度/月（Pro）
上下文200K tokens
额度持久约 2 倍
Enterprise $60/用户有 1,000 额度/月。RTK 压缩 CLI 输出，让额度更持久。
Gemini CLI Terminal
价格免费 — 按 token 付费
限制每天 1,000 请求，每分钟 60 请求（免费）
上下文1M tokens
token 账单减少约 70%
免费层很慷慨但仍有速率限制。RTK 压缩 89%，减少账单或释放速率限制空间。
Aider Terminal
价格免费 + API 费用（$5-300+/月）
限制按 API 提供商
上下文按模型（最高 200K）
API 成本减少约 70%
自带 API 密钥—按 token 付费。RTK 压缩每个命令输出，直接减少约 70% API 费用。
GitHub Copilot IDE
价格免费 — $39/月（Pro+）
限制每月 50-1,500 高级请求
上下文按模型（最高 200K）
更好的上下文质量
Enterprise $39/用户：1,000 高级请求/月。RTK 保持终端输出精简，高级请求携带更多有用上下文。
Cline / Roo VS Code
价格免费 + API 费用（$0-500+/月）
限制按 API 提供商
上下文按模型（最高 200K）
API 成本减少约 70%
无工具端限制。RTK 压缩 89%，直接减少账单和上下文溢出。

价格于 2026 年 2 月验证。RTK 节省基于 2,900+ 个真实命令的平均 89% 压缩率。

即将推出

RTK Cloud

对团队 AI 编程成本的可见性和控制。了解浪费。解决它。

Token 分析

按开发者、项目、工具的仪表盘

团队节省报告

"你的团队本月节省了 $4,200"

速率限制警报

监控和智能通知

企业控制

SSO、审计日志、合规

开源免费。团队从 $15/开发者/月 起。

0 团队在等待列表中

无垃圾邮件。上线时发一封邮件。

Viking slamming axe into ground creating green energy shockwave
// 快速开始

30 秒开始

安装，激活自动重写钩子，每个命令自动压缩。

快速安装

Linux 和 macOS 一行命令

curl -fsSL https://raw.githubusercontent.com/rtk-ai/rtk/refs/heads/master/install.sh | sh

Via Homebrew

macOS & Linux

brew install rtk
brew upgrade rtk

预编译二进制

macOS、Linux、Windows

从 Releases 下载 →

然后激活自动重写钩子

rtk init --claude-code

在 Claude Code settings.json 中安装 PreToolUse 钩子——每个 Bash 调用都会自动重写。 完整安装指南 →

1 curl ... | sh
2 rtk init --claude-code
3 rtk gain
Viking raising axe in triumphant war cry
// 发布

你的 AI 不需要
读完这些。

安装 rtk。更好的代码，更长的会话，更低的成本。

安装 rtk GitHub 上加星

Frequently asked questions

What is RTK (Rust Token Killer)?

RTK is an open-source CLI tool that compresses command outputs before they reach the AI context window. It reduces token usage by 60-90% with zero configuration changes, enabling longer AI coding sessions and lower API costs. RTK is written in Rust, MIT licensed, and works transparently with Claude Code, Cursor, and any terminal-based AI assistant.

How many tokens does RTK actually save?

Based on measurements across 2,900+ real-world commands, RTK removes an average of 89% of CLI output noise. Command-level savings: cargo test (91.8% savings), git status (80.8%), find (78.3%), grep (49.5%). A developer who ran 15,720 commands saved 138 million tokens over several weeks, tracked live via rtk gain.

Is RTK free? Are there usage limits?

RTK is completely free. It is open source under the MIT license, with source code available on GitHub at github.com/rtk-ai/rtk. There are no usage limits, no API keys required, no telemetry, and no accounts. RTK Cloud (waitlist) will offer additional team features.

Which AI coding tools and commands does RTK support?

RTK works with Claude Code (Anthropic), Cursor, Gemini CLI, Aider, and any AI assistant that reads terminal output. It supports all major CLI commands: cargo test, pytest, go test, git diff, git status, git log, grep, find, ls, pnpm list, tsc, eslint, prisma, docker, kubectl, and more.

How does RTK work without changing my workflow?

Running rtk init --global installs a PreToolUse hook in Claude Code that automatically rewrites Bash commands to rtk equivalents at the proxy layer. You continue using your normal commands. RTK intercepts and compresses the output before it enters the context window, with no manual changes to prompts or workflows needed.

Does RTK affect code quality or AI reasoning accuracy?

No. RTK removes verbose boilerplate and repetitive output noise, not meaningful content. Test failures, error messages, diffs, and stack traces are preserved in full. The AI receives the same essential information with 89% less noise, which typically improves reasoning quality by reducing context pollution.