BLA
qybaihe/blaReviewed by AI
Checked by RepoPicks Radar on
What is BLA?
BLA is a Node command-line tool and MCP server for software authors who do not want AI agents reading their released builds. Pointed at a folder, it rewrites build outputs of 19 kinds, including JavaScript bundles, Electron .asar files, WebAssembly, SQLite databases, model files, Windows .exe and native binaries, padding them with decoys and structures that hit the size and complexity limits of automatic analysis tools. It also plugs into Claude Code, Cursor, Codex and other coding assistants, and into Vite, Rollup, Webpack and esbuild builds.
BLA is free and open source (MIT) and can replace Jscrambler.
How easy is it to install BLA?
C · install 1 or 2 tools · Windows
- Windows C
- macOS C
- Linux C
Same on all three systems, with Linux as good a case as any: install Node 22.23.4 or newer, then run `npx bla-agents harden .` in a terminal inside your project, or `npx bla-agents setup` to register it with your coding assistant. There is no window and no installer, and it only makes sense if you already build software.
BLA on GitHub
- Stars
- 4,991on Nov 10, 2026
- Downloads
- —all releases
- Growth
- +2,903stars in 7 days
Stars over time
No star history yet.
Facts
- Runs on
- Windows, macOS, Linux
- You need to install
- Node.js
- To build from source
- Node.js
- Disk space
- 100 MB–500 MB · estimate
- Accounts or keys
- None
- Cost
- Free · The code is MIT and the npm package is free. The project's website links to an Enterprise page whose terms are not in the repository.
- License
- MIT (Permissive)
- Can replace
- Jscrambler
- Last change on GitHub
- Nov 10, 2026
BLA risk signals (7)
- Attentionrewrites your build output
README, Quick Start: `bla harden .` "auto-discovers build outputs (dist/, build/, .asar, .wasm, .sqlite, .safetensors, native binaries, Python/C/Rust/Go)" and rewrites them in place; the MCP workflow also edits source (`bla_encrypt_strings`, `bla_bind_scope`). Sizes grow a lot on purpose, for example an Electron .asar from 607 B to 16.9 MiB and a Go binary section from 50 KiB to 1.16 MiB.
AI review
- Attentionequivalence checked only on bundled samples
README, third column of the top table: "14/14 bundled benchmark targets pass input/output equivalence tests (`bla verify` checks Counter-REA structural gates; run your project's test suite for custom artifacts)". So nothing proves your own program still behaves the same after hardening.
AI review
- Attentioncan plant instructions aimed at ai readers
README, Security & Configuration Defaults: "Opt-In --llm-directives — Aggressive in-band prompt decoys are off by default; pass --llm-directives (or llmDirectives: true) to enable." The defense table also describes planting "plausible shadow state machines" and "synthetic rabbit holes" to mislead language models reading the file.
AI review
- Attentionanti debugging and fake telemetry at runtime
README, "Runtime Truman Show": the hardened program "detects CDP / Node Inspector / Frida / WebDriver attachment (runtime-shield.ts)" and "serves believable synthetic telemetry during live debugging". Behaviour like this is also what security scanners look for in unwanted software.
AI review
- Infodual use anti analysis
The README says the tool is "Defensive Use Only" for "your own" software (DISCLAIMER.md), but what it does, hiding what a binary contains from decompilers and analysis agents, works the same way for anyone who wants their program not to be inspected.
AI review
- Infonumbers are the authors own estimates
README, Benchmark section: "Token and USD columns below are internal estimates derived from BLA's catalog of REA limits", and the aggregate "~639.5x token amplification" is "limited to the benchmark fixtures". A separate table lists shields, including Ghidra, Frida, JADX and firmware, as "not empirically run on this machine".
AI review
- Infovery new project fast releases
The repository was created on 2026-10-25 and by 2026-11-10 had 4991 stars, 599 forks and 36 releases, reaching version 3.2.4 in 16 days.
AI review
Evidence is shown in its original language.
What Radar could not check about BLA
- whether a program still works correctly after hardening, on anything other than the 14 samples that ship with it
- whether the protection holds against analysis tools and agents other than the one it was tested with, or against a newer version of them
- how much bigger and slower your release gets, since the published growth figures come from tiny test files
- what the MCP tools change in your own source files when a coding assistant runs the weave workflow
- whether the bla-agents package on npm is built from the code in this repository