EgoAI is the reasoning and coordination module of a modular synthetic mind. Inspired by Freudian and cognitive‑neuroscience models, it acts as the Ego - the center of goal selection, attention, and memory control.
Unlike end‑to‑end LLM systems, EgoAI separates perception, memory, and logic. This yields interpretability, modularity, and human‑like reasoning without neural networks.
Architecture (high‑level)
Synthmind (organs) → Token → EgoAI ← Sentience (DSL) ← Inception (Memory)| Module | Role |
|---|---|
| Synthmind | Sensory input system (e.g., vision, audio). Emits perceptual tokens. |
| EgoAI | Control logic. Interprets inputs, uses Sentience DSL to read/write memory, plans and executes actions. |
| Sentience | DSL for declarative memory access and agent goals. |
| Inception | Cortex & memory store: STM, LTM, latent memory; queried via DSL. |
EgoAI is stateless; it interprets perception, manages memory, and executes goals.
Quick Start
Make sure supporting services are running: inception-core (Rust memory server @ :8080) and sentience (Rust REPL/DSL) - optional recognizer (vision/audio via Flask).
- Clone & run
git clone https://github.com/nbursa/EgoAI
cd EgoAI
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
python cmd/main.py- Run recognizer (optional)
cd recognizer
python recognizer.py- Configure endpoints
Create
.envwith memory/DSL endpoints:
INCEPTION_URL=http://localhost:8080
SENTIENCE_URL=http://localhost:8777Core Concepts
- Token - the atomic unit of perception. Carries content and modality (vision, audio, text).
- MemoryToken - Sentience → Inception. Stores subject, relation, object, and tags.
- Agent Goal - expressed in Sentience DSL; EgoAI interprets and executes it.
- Memory - stratified: STM (short‑term), LTM (episodic/semantic), and latent. Queried via Sentience instructions.
Memory Architecture (doc)
EgoAI uses a layered memory system designed for interpretability:
- Observation Layer (input translation)
- STM (short‑term buffer)
- Episodic / Semantic / Procedural / Reflective memories
- MemoryStore access layer with indexing & write/retrieval protocol
See the full spec with examples and protocols here: /docs/memory_architecture.md
Why EgoAI?
- Interpretability by design: explicit tokens and DSL instead of opaque embeddings.
- Modularity: swap perception, memory engines, or planning policies independently.
- Deterministic control loop: reproducible behavior; easy to test.
- Human‑like workflow: separates observation → memory → reasoning → action.
Related Architectures
- ACT‑R • Soar • OpenCog