Hi everyone!
I recently experimented with a new approach: using a plain .txt file (TXT OS) to upgrade the reasoning, memory, and safety of any LLM (like ChatGPT, Claude, Gemini, etc) — no installs, no code, just copy-paste.
What does it do?
- Adds semantic memory (remembers your logic, not just tokens)
- Shields against hallucinations and boundary errors (ΔS/semantic constraints)
- Fully open source (MIT), no telemetry, no tracking
- Works with any model that supports prompts
I built and released a series of .txt-based tools:
- TXT OS — reasoning “operating system” in a text file
- Bla Bla Bla — Q&A engine for wild/creative semantic questions
- Blur Blur Blur — text-to-image with nearly zero hallucination
- Blow Blow Blow — reasoning-based text adventure/game
Benchmarks:
+22.4% semantic accuracy, +42.1% reasoning success, ×3.6 long-chain stability (GSM8K, Truthful-QA, temp 0.2)
If you’re interested, everything is free/open source here:
WFGY/OS at main · onestardao/WFGY · GitHub
Would love to hear feedback, questions, or ideas for new modules!
(And happy to help beginners or advanced users alike.)
— PSBigBig / Purple Star
