Every week, viral X threads promise "free tools agencies charge thousands for." We open the links, check the stars, license and README, and give you the catch. Updated weekly. Week of Oct 14, 2026.
| Repo | Stars | License | Verdict | What it does | The catch |
|---|---|---|---|---|---|
| REA (morluto/rea) | 249K | MIT | Real | Reverse-engineering MCP that routes binaries, .NET, APKs and Electron apps to Ghidra/JADX and similar engines. | The hype says 'free paid apps'. Using it to bypass licenses is piracy. It's legit for auditing your own or authorized software. |
| DeepSeek Harness (dsh) | 250K | MIT | Real, preview | Plugin-based agent harness with a local web UI (npx @deepseek-ai/dsh web). | Developer preview with breaking changes. Read the safety notice first. |
| Herdr | 44K | Apache-2.0 | Real | Runtime that runs several coding-agent sessions in tabs. | Useful if you run several agents. Saves no money by itself. |
| Humanizer | 57K | MIT | Real | A skill that strips robotic AI-writing patterns from text. | Edits style only. It doesn't make thin content good. |
| Caveman | 112K | Apache-2.0 | Real | Makes agent replies short and blunt, which cuts output tokens. | Read its hooks before running the installer. |
| Defuddle | 9.7K | MIT | Real | Turns cluttered web pages into clean markdown for agents. | Small and safe. Tested: works. |
| Agent-Reach | 97K | MIT | Real | Lets an agent research YouTube, X, Reddit and GitHub. | Some of its sources may need your own cookies or keys. Don't hand over your logins. |
| Codebase Memory MCP | 47K | MIT | Real | Keeps a persistent map of a large codebase. | Only worth it for big repos. |
| TencentDB Agent Memory | 29K | Custom (check) | Real | Local layered memory (L0 chat, L1 atom, L2 scenario, L3 persona) on SQLite. | 'Killed vector DBs' is hype. Check the license before any commercial use. |
| OpenMontage | 68K | AGPL-3.0 | Real | Agent video studio: pipelines for clips, avatars and localization. | AGPL. Heavy setup, and many of its providers are paid APIs. |
| OpenShorts | 140 | MIT | Real, small | Long video in, vertical captioned clips out. | Tiny project. The viral post overstates it. |
| iFixAi | 28K | Apache-2.0 | Real | Grades an agent A-F across 60 checks. | Telemetry is ON by default, and the vendor sells paid audits. Turn telemetry off. |
| xtool | 11K | MIT | Real | Builds and signs iOS apps without a Mac. | Still needs an Apple developer account ($99/yr) to ship. |
| PentAGI | 30K | MIT | Real | Autonomous pentest agents. | Only for systems you own. No income angle. |
| Decepticon | 6K | Apache-2.0 | Real | Red-team agent swarm. | The 98% benchmark claim is the vendor's own number. Authorized targets only. |
| Whisper | Huge | MIT | Real | Local speech-to-text. | Classic. Use faster-whisper on CPU. |
| Aider | Huge | Apache-2.0 | Real | Git-aware AI pair programmer that works with Ollama. | Local models are weaker at multi-file edits. |
| DSPy | Huge | MIT | Real | Optimizes prompts against a metric instead of guessing. | Needs a small labeled set to be useful. |
| crawl4ai | Huge | Apache-2.0 | Real | LLM-friendly web crawler. | Respect robots.txt and site terms. |
| Ai-jailbreak | ? | ? | Avoid | Collection of jailbreak prompts. | No legit business use. Account-ban risk. |
Not affiliated with any repo listed. Verify licenses yourself before commercial use.