Boost is a CLI wrapper that makes every command your coding agents run faster, quieter, and cheaper in tokens. You prefix any command with `boost` (for example, `boost docker build`) and Boost runs the real tool transparently while replacing its noisy output with a compact, agent-friendly summary.

Which coding agents does Boost support?

Cursor, Claude Code, Codex CLI, and GitHub Copilot — Boost works behind the scenes with the agents you already use.

How does Boost make commands faster and quieter?

Boost runs the underlying command (docker, npm, go, gh, git, vite, bun, and 100+ more) and applies a per-command filter that strips progress bars, repeated chatter, and ANSI noise. For supported tools it also adds intelligent caching. Typical token reductions are 60–90%.

Yes. The token-saving features in the Boost CLI are free, and the core will stay free forever — no signup, no credit card, no trial timer. Later on, some advanced, enterprise-grade capabilities (think team-scale governance and managed acceleration) may be paid, but the CLI that makes your agent runs leaner and cheaper stays free. We will always be upfront about what is free and what is not.

We built Boost to fix our own bloated, expensive agent runs — and it turned out to help everyone. Keeping the token-saving CLI free is how JFrog gives something genuinely useful back to the developer community: if it saves you tokens and money, that is the whole point. Any future paid enterprise features are there to fund the free tier for large teams, not to gate the basics.

Who should use Boost?

Boost is built for everyone from startups to enterprises: solo developers using coding agents, teams trying to keep agent context under control, and organizations that want leaner, cheaper agent runs without changing the tools they already use.

How does Boost handle my data and privacy?

Your code, logs, and terminal output stay on your machine — all filtering and compaction happens locally and is never uploaded. The only thing Boost sends is lightweight telemetry about how many tokens it saved. Our philosophy is simple: we learn from how Boost performs in the real world so we can keep making it faster and smarter for the people who use it, without ever touching what is private to you.

Boost CLI: Cut Coding-Agent Token Waste

What builders are saying

Posts from people using Boost — tap a card to read the original.

Roy

Integrating boost.jfrog.com into our pipeline has been a massive win for my team and I. We’ve officially saved over 79 million tokens while keeping our AI-assisted workflows fast, precise, and context-aware. If you’re scaling LLM usage in production, this is an absolute must-use tool. 🚀

hiroppy | iiba

@about_hiroppy

jfrogもAIトークン抑えるツール出しとるやん。 jfrog/boost: Save tokens. Maximize context, Safely

Yuval Lewit

yuval-lewit

Been using JFrog Boost for the past few weeks and wanted to share real numbers instead of just vibes. 13.5M tokens saved over the last 26 days. CLI filtering alone accounted for 3.1M of that. It's not a black box — everything stays local, and every compression is reversible.

Roy

Integrating boost.jfrog.com into our pipeline has been a massive win for my team and I. We’ve officially saved over 79 million tokens while keeping our AI-assisted workflows fast, precise, and context-aware. If you’re scaling LLM usage in production, this is an absolute must-use tool. 🚀

hiroppy | iiba

@about_hiroppy

jfrogもAIトークン抑えるツール出しとるやん。 jfrog/boost: Save tokens. Maximize context, Safely

Yuval Lewit

yuval-lewit

Been using JFrog Boost for the past few weeks and wanted to share real numbers instead of just vibes. 13.5M tokens saved over the last 26 days. CLI filtering alone accounted for 3.1M of that. It's not a black box — everything stays local, and every compression is reversible.

Liam

@ottleyai

AI coding agents burn a surprising amount of context on stuff they probably never needed to read. Test logs. CLI output. Repeated code searches. JFrog built a free CLI called Boost that compresses that noise before it reaches the model. One test went from 232 lines → 22 while keeping the actual failure intact.

Itai Ben Zeev

itai-ben-zeev-ar

87% פחות טוקנים על אותה פקודת go test. זה מה שיצא לי כשמדדתי בעצמי. Boost זה CLI חינמי שעוטף פקודות כמו npm ci, docker build ו-pytest, ומסנן לסוכן את כל הרעש מסביב לפני שהוא שורף לנו טוקנים.

Yehonatan Dahan

yehonatan-dahan-yd

If you're building with LLMs at scale, token optimization and context management are everything. Recently, I've managed to save over 30M+ tokens a month using boost.jfrog.com.

Liam

@ottleyai

AI coding agents burn a surprising amount of context on stuff they probably never needed to read. Test logs. CLI output. Repeated code searches. JFrog built a free CLI called Boost that compresses that noise before it reaches the model. One test went from 232 lines → 22 while keeping the actual failure intact.

Itai Ben Zeev

itai-ben-zeev-ar

87% פחות טוקנים על אותה פקודת go test. זה מה שיצא לי כשמדדתי בעצמי. Boost זה CLI חינמי שעוטף פקודות כמו npm ci, docker build ו-pytest, ומסנן לסוכן את כל הרעש מסביב לפני שהוא שורף לנו טוקנים.

Yehonatan Dahan

yehonatan-dahan-yd

If you're building with LLMs at scale, token optimization and context management are everything. Recently, I've managed to save over 30M+ tokens a month using boost.jfrog.com.

桃芭蕉実

@momoandbanana22

おお～token節約、動いてる～♪ 使ってみたのは、これです → jfrog/boost

Alexey Geht

alexeygeht

I like tools like this. JFrogg released Boost for Claude Code, and it solves a very real problem: too much useless terminal noise getting shoved into the context window. Keep the real signal, like the failing test, the error, the summary. Drop the spammy output that burns tokens and adds nothing.

桃芭蕉実

@momoandbanana22

おお～token節約、動いてる～♪ 使ってみたのは、これです → jfrog/boost

Alexey Geht

alexeygeht

I like tools like this. JFrogg released Boost for Claude Code, and it solves a very real problem: too much useless terminal noise getting shoved into the context window. Keep the real signal, like the failing test, the error, the summary. Drop the spammy output that burns tokens and adds nothing.

たいしょー

@N_Taisho

JFrogが、AIコーディングエージェント向けのログ圧縮ツールBoostを公開してる。npm ciでは約9,800トークンの出力を約640トークンまで圧縮しつつ、エラーなど必要な情報は残す。ベンチマークではタスク成功率を維持しながらコストを約12%削減。

Dor Shaboy

AWS

I wasted thousands of tokens last month without knowing it. Your actual message? ~1% of the bill. The other 99% is repeated context the model already had. I started using JFrog Boost a few weeks ago. It compresses noisy tool output, drops unused tool schemas, optimizes file reads, and filters MCP calls — all before they hit the context window.

たいしょー

@N_Taisho

JFrogが、AIコーディングエージェント向けのログ圧縮ツールBoostを公開してる。npm ciでは約9,800トークンの出力を約640トークンまで圧縮しつつ、エラーなど必要な情報は残す。ベンチマークではタスク成功率を維持しながらコストを約12%削減。

Dor Shaboy

AWS

I wasted thousands of tokens last month without knowing it. Your actual message? ~1% of the bill. The other 99% is repeated context the model already had. I started using JFrog Boost a few weeks ago. It compresses noisy tool output, drops unused tool schemas, optimizes file reads, and filters MCP calls — all before they hit the context window.

Terminal Bench 2.0

Same 89 tasks. Same success rate.Lower cost with boost

Cost per task, measured on Terminal-Bench 2.0.

Read the full report

Cost per task · lower is better · Δ vs baseline

Baseline$0.319—

RTK$0.339+6.3%

Boost$0.276-13.5%

Read the full report

Enterprise backed. Platform ready.

The Boost CLI is free — not a paid JFrog platform SKU.

Backed by JFrog

Built by the team behind the platform thousands of teams trust to secure and ship software.

why-boost — workspace

Explorer

why-boost

enterprise-security.md×

1

2

3

4

5

6

7

# Enterprise security

DevSecOps DNA from the platform trusted

to secure software supply chains.

- Secure by default

- Backed by JFrog's security platform

Questions about Boost

Everything you might want to know before you install.

# select a question

boost — faq

$ boost

New version of Boost released— More reliable BoostGraph on Windows

Save Tokens

What builders are saying

CLI Filters

Boost Graph

MCP Filter

File Optimization

Same 89 tasks. Same success rate.Lower cost with boost

boost retrieve

Enterprise backed. Platform ready.

Backed by JFrog

Questions about Boost

Take control of your token spend.