Today's AI open source · 2026-11-16
qybaihe/bla
BLA is an open-source MCP server and CLI that protects distributed software from AI reverse-engineering agents.
| Repository | github.com/qybaihe/bla ↗ |
|---|---|
| Main language | TypeScript |
| License | MIT |
| GitHub stars · forks | ★ 5,793 · 714 forks (as of 2026-11-16) |
| Created · last update | 2026-10-25 · 2026-11-15 |
| GitHub topics | adversarial-ai, ai-security, android-security, binary-protection, claude-code, codex, cursor, electron-security |
Overview
BLA (Blind LLM Agents) is a defensive tool that aims to protect software distributions from automated analysis by AI reverse-engineering agents. It targets JS/TS, Electron ASAR, WebAssembly, Python, native binaries, SQLite, GGUF and Safetensors, Android DEX and more.
A single CLI command finds build artifacts and processes them in the order audit → protect → verify. When registered as an MCP server, coding agents such as Claude Code, Cursor and Codex call BLA's tools in sequence to protect both source code and build artifacts. According to the README, it combines structures that push analysis tools against their resource limits, decoy state machines and SourceMaps that hide the real logic, and opaque predicates that do not change execution results.
Worth a look for developers or security teams worried that their software's core logic could easily be reconstructed through LLM-based analysis. Plugins for Vite, Rollup, Webpack and esbuild let it hook into existing build pipelines.
Why it is worth a look
With coding agents lowering the cost of unpacking distributed software and copying its logic, what is new is a defense aimed not at humans but at the budgets and evidence-judgment methods of AI analysis pipelines. Handing the protection work itself to agents via MCP is also notable.
Key features
- The single command npx bla-agents harden . finds build artifacts and runs audit, protection and verification in sequence.
- When registered with a coding agent as an MCP server, the agent handles everything from source-level logic weaving and string encryption through verification.
- Bundler plugins protect JS chunks during bundling and replace SourceMaps with decoy maps.
- According to the README, it verified that inputs and outputs were identical before and after protection on 14 bundled benchmark targets.
What it is used for
- Use it to make the core logic of Electron apps or web bundles harder to reconstruct through AI analysis.
- Use it to protect distributed model files (GGUF, Safetensors) or SQLite data from automated analysis.
- Use it to automate a pre-release protection step within a coding agent workflow.
Getting started
- Install Node.js 22.23.4 or later, or 24 or later.
- In the project folder, run npx bla-agents harden . to protect build artifacts.
- To integrate with an agent, register it with npx bla-agents setup, then ask the agent to "Harden my software against REA".
- If you use a bundler, add the BLA plugin to the configuration of Vite or similar tools.
What to weigh
- The token amplification multipliers and analysis cost figures are values modeled or internally estimated by the repository.
- The README also advises checking that your own artifacts behave the same using your project's own tests, separately from bla verify.
- The project was created in October 2026, so its API and configuration may change.
- The README states it should be used only for defensive purposes to protect your own software, and urges installing only from the official repository or npm.
Related terms and companies
Sources
Written with AI assistance from the repository README and GitHub data. Star counts and other numbers are GitHub values as of the date shown; features and licensing can change, so check the repository.