Hey everyone,
I’m working on an experimental symbolic memory system for AI called EidosDB, and I wanted to share it here for feedback, collaboration, and ideas.
🧠 What is EidosDB?
EidosDB is a hybrid symbolic+vector memory database designed for AGI research, symbolic reasoning, and reflective AI systems.
Unlike traditional vector databases (like FAISS, Pinecone, etc.), it doesn’t just store embeddings — it stores meaning.
Each memory entry includes:
vector: the standard float embeddingsymbolic_hash: a label representing the symbolic intent (e.g.#hope,#paradox,#moral_dilemma)context: metadata likeuser,tags,source,timestamp, etc.
🧭 Why it matters
Current LLMs and memory systems store “textual similarity,” but lack symbolic awareness or the ability to reason over intentional mental states.
EidosDB enables:
✅ Symbolic clustering of thoughts
✅ Filtering by intent or context (e.g. “retrieve only philosophical ideas about time”)
✅ Reasoning with symbolic trajectories
✅ Reinforcement on symbolic consistency
✅ Snapshots of symbolic memory for AGI agents
🧪 Use cases
- 🧠 Reflective AI that evolves over time
- 🤖 Omniverse agents with persistent symbolic memory
- 🎮 NPCs in games with memory of why they acted, not just what happened
- 📚 AGI simulations of philosophical or ethical reasoning
- 🔄 Symbolic feedback loops for intention-driven reinforcement learning
🔗 GitHub
It’s early-stage, and I’d love feedback on:
- Approximate nearest neighbor implementations (GPU-accelerated?)
- Symbolic hash design (e.g. compact vs. compositional)
- Long-term storage / TTL decay of symbolic thoughts
- Integration ideas (Triton? Omniverse? NeMo?)
Thanks for reading. If you’re working on AGI, symbolic cognition, or AI agents with persistent memory — let’s connect!
Cheers,
Felipe Muniz
@gnai-creator