AI Offline Chat ADV is a modern, fully autonomous, and offline implementation of the legendary ELIZA—the world’s first artificial intelligence therapist, originally created at MIT in 1966. Re-engineered from the ground up for the M5Stack Cardputer, this firmware uses advanced C++ optimizations, JIT-compiled Regular Expressions (RegEx), and hardware-level memory protection to run a pure NLP (Natural Language Processing) engine directly on the ESP32-S3 chip, without requiring Wi-Fi or cloud APIs.
Important
Source Code Status: This project is proprietary. The source code is private.
Distribution: Binary (.bin) only via the Releases tab.
This is NOT ChatGPT. Please understand what you are installing:
- No Cloud, No Servers: The AI runs 100% locally on the microcontroller's limited SRAM.
- Not an Encyclopedia: If you ask her "What is the capital of France?" or "Write code for me", she will not answer. She has no external knowledge base.
- The Rogerian Therapist: ELIZA is a psychological simulation. She uses a technique called "active listening." Her purpose is to analyze your grammar, reflect your statements back as questions, and encourage you to talk about your feelings, relationships, and problems.
- The Illusion of Understanding: She doesn't "think" in neural weights; she processes language via complex linguistic rules. The magic happens when you engage with her as if she were a real listener.
- True Offline NLP Engine: Processes Natural Language offline using a custom C++ parsing core, bypassing the need for heavy LLM models.
- JIT-Compiled Regex System: Employs Just-In-Time compilation for over 30 complex psychological trigger rules. The engine dynamically scans input and instantly flushes memory to prevent Heap fragmentation.
- Hardware-Level Stack Protection: Embedded C++
<regex>engines are notoriously dangerous on microcontrollers. This firmware features a custom sanitizer that truncates infinite quantifiers (.*), physically preventing RTOS Stack Overflows during deep recursive sentence analysis. - Long-Term SD Memory: ELIZA remembers. She saves your name, family members, specific objects you mention, the overall "emotional tone" of your last session, and your system volume preferences to the SD card.
- Open M5Burner.
- Search for
AI Offline Chat ADVorZeloksa. - Select version V1.1.
- Burn to your M5Stack Cardputer.
- Go to the [Releases] tab of this repository.
- Download the
ELIZA1.1.bin(full 4B flash dump) or the standard app bin. - Flash to your Cardputer using M5Burner (Local File) or esptool.
- [ Full Keyboard ]: Type your thoughts normally.
- [ ENTER ]: Send message to ELIZA.
- [ DEL ]: Backspace / Delete character.
- [ ; / . ]: Scroll chat history Up/Down (if the chat exceeds screen height).
- [ - / _ ]: Decrease typing volume / Mute.
- [ = / + ]: Increase typing volume.
- [ ESC / ` ]: Open System Menu (Start New Session / Wipe Memory).
For the best experience, type in full English sentences. Use pronouns (I, My, You) and express states or emotions.
- Bad Input: "apple" or "hello bot"
- Good Input: "I feel really anxious about my new job today." or "My mother never listens to me."
An SD Card must be inserted for the long-term memory to function. The system creates a tiny profile (/eliza_v7.txt) where it stores:
- Named Entities: Your name and specific objects you care about.
- Emotional Context: If your session focuses on grief (
TOPIC_GRIEF), the UI will adapt (green text, lower tone frequency), and ELIZA will remember your state across power cycles. - Color-Coded History: The chat history is saved along with its specific emotional color coding.
- Hardware States: Your preferred typing volume level is permanently saved.
When you submit a complex sentence, the ESP32-S3 must compile and test multiple Regex patterns. You will see an ELIZA: ... indicator. This is not a fake animation; it represents the actual CPU blocking time required for the NLP engine to calculate its response before rendering it via a typewriter effect.
If you wish to erase your psychological profile and start fresh, press ESC or ` to open the System Menu. From there, you can select the option to wipe the SD card memory and reset her context variables.
- Added: Adjustable typing volume using
-and=keys (includes full mute option). - Added: Volume level state is now persistently saved to the SD card profile.
- Fixed: Eliminated screen tearing during the boot animation by implementing dynamic double-buffering (M5Canvas).
Support the development of advanced offline tools and engines for the Cardputer ecosystem:
Developed by Engineer Zeloksa. Strictly optimized for Cardputer ADV.