Skip to content
Open Source · MIT · Rust 50.3k stars

Tu agente IA se ahoga
en ruido del CLI. Arréglalo.

rtk comprime las salidas de comandos antes de que lleguen a la ventana de contexto. Mejor razonamiento. Sesiones más largas. Menores costos.

89% ruido eliminado (prom.)
3x sesiones más largas
30+ comandos
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).

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
// 01 — el problema
Viking bracing against a tidal wave of tokens

El problema del coding con IA hoy

Cada comando que ejecuta tu agente contamina la ventana de contexto con ruido. Esto es lo que te cuesta.

Contaminación del contexto

Tu ventana de 200K tokens no es infinita. Cuando cargo test vuelca 5.000 tokens de boilerplate, son 5.000 tokens menos para razonar sobre tu código.

context_quality: degraded ▼

Sesiones demasiado cortas

El contexto se desborda, el agente reinicia, pierdes el hilo. Con tarifas planas, alcanzas los límites un 40% más rápido.

session_remaining: 32% ▼

Costos que explotan

Con pago por token (API, Gemini CLI, Aider), el 70% de tu factura es ruido que el LLM no necesita. Un equipo de 10 desperdicia ~$1.750/mes.

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

Mira la diferencia

Salidas reales, ahorros reales. Comparación lado a lado con comandos reales.

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

Ahorros reales

Resultado real de un desarrollador satisfecho con rtk gain.

Feedback de un desarrollador

Tras unas semanas de uso diario: 15.720 comandos, 138M tokens ahorrados. Ejecuta rtk gain para ver los tuyos.

88.9% efficiency
rtk gain dashboard showing 88.9% efficiency

Desglose detallado

Estadísticas diarias, semanales y mensuales por comando. Rastrea tus ahorros.

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

Ninguna herramienta IA ofrece uso ilimitado. Cada token cuenta.

Incluso a $200/mo, todas tienen límites. RTK comprime el ruido CLI para que tus límites rindan más.

Una sesión típica de 2h con un agente IA:

~60 comandos CLI ejecutados
~210K tokens de ruido CLI
~23K con RTK (89% menos)

Sin RTK, las salidas CLI solas pueden desbordar una ventana de 200K tokens. Basado en prom. 3.500 tokens/comando medidos en 2.900+ comandos reales.

Claude Code Terminal
Precio$20 — $200/mo
Límites~45 msgs/5h (Pro), 5-20x en Max
Contexto200K tokens
Sesiones ~3x más largas
Incluso Max $200/mo (20x Pro) tiene límites semanales. La cuota se reinicia cada 5h. RTK comprime salidas CLI un 89% en prom., tu cuota se extiende ~3x.
Cursor IDE
Precio$20 — $200/mo
Límites$20 créditos/mo (Pro), ~225 reqs Claude
ContextoHasta 200K (modo Max)
Créditos rinden ~2x más
Incluso Ultra $200/mo está limitado. RTK comprime salidas CLI para arrancar más limpio.
OpenAI Codex Agent
Precio$20/mo (Plus) — $200/mo (Pro)
Límites30-1.500 msgs/5h según plan
Contexto192K tokens
Más iteraciones por cap
Incluido con planes ChatGPT. Pro $200/mo tope de 1.500 msgs/5h. RTK los comprime para más iteraciones.
Windsurf IDE
Precio$15 — $60/mo
Límites500 créditos/mo (Pro)
Contexto200K tokens
Créditos duran ~2x más
Enterprise $60/user obtiene 1.000 créditos/mo. RTK comprime salidas CLI para que cada interacción cueste menos tokens.
Gemini CLI Terminal
PrecioGratis — por token
Límites1.000 req/día, 60 req/min (gratis)
Contexto1M tokens
~70% menos en factura
El tier gratis es generoso (1.000 req/día) pero limitado. RTK comprime salidas CLI un 89% en prom., reduciendo tu factura.
Aider Terminal
PrecioGratis + costos API ($5-300+/mo)
LímitesSegún proveedor API
ContextoSegún modelo (hasta 200K)
~70% menos costo API
Clave API propia — pagas por token. RTK comprime cada salida de comando antes de llegar al modelo, reduciendo tu factura API un ~70%.
GitHub Copilot IDE
PrecioGratis — $39/mo (Pro+)
Límites50-1.500 reqs premium/mo
ContextoSegún modelo (hasta 200K)
Mejor calidad de contexto
Enterprise $39/user: 1.000 reqs premium/mo. RTK mantiene salidas de terminal ligeras para que las reqs premium lleven más contexto útil.
Cline / Roo VS Code
PrecioGratis + costos API ($0-500+/mo)
LímitesSegún proveedor API
ContextoSegún modelo (hasta 200K)
~70% menos costo API
Sin límite del lado herramienta. RTK comprime cada salida un 89% en prom., reduciendo la factura.

Precios verificados feb. 2026. Límites varían según uso y plan. Ahorros RTK basados en 89% de compresión promedio en 2.900+ comandos reales.

Próximamente

RTK Cloud

Visibilidad y control sobre los costos de IA de tu equipo. Sabe qué se desperdicia. Corrígelo.

Analítica de tokens

Dashboard por dev, por proyecto, por herramienta

Informes de ahorro

"Tu equipo ahorró $4.200 este mes"

Alertas de límites

Monitoreo y notificaciones inteligentes

Controles empresariales

SSO, logs de auditoría, cumplimiento

Gratis para open-source. Equipos desde $15/dev/mes.

0 equipos en lista de espera

Sin spam. Un solo email al lanzamiento.

Viking slamming axe into ground creating green energy shockwave
// empezar

Empieza en 30 segundos

Instala, activa el hook auto-rewrite, y cada comando se comprime automáticamente.

Instalación rápida

Una línea para Linux y macOS

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

Binarios pre-compilados

macOS, Linux, Windows

Luego activa el hook auto-rewrite

rtk init --claude-code

Instala hook PreToolUse en Claude Code settings.json — cada llamada Bash se reescribe automáticamente. Guía completa →

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

Tu IA no necesita
leer todo eso.

Instala rtk. Mejor código, sesiones más largas, menores costos.

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.