The memory layer for the next generation of intelligent agents.
- Official Docs: docs.buildpersona.ai
- Deep Dive: The Philosophy of Persona (saxenauts.io)
Memory in AI today is often treated as simple storage—a static database of vectors or JSON blobs. Tools like Mem0 and Graphiti provide excellent infrastructure for this, handling the "storage" aspect efficiently.
But true intelligence requires more than storage. It requires a dynamic, living system that evolves with the user.
Persona aims to be that intelligent layer.
- Beyond Storage: Just as OpenRouter revolutionized model access by adding intelligence on top of raw APIs, Persona adds intelligence on top of raw memory.
- Memetic Organism: We view the user's digital identity not as a table of rows, but as a living graph—a "memetic organism" that grows, forgets, and reinforces connections based on interaction.
- Intelligent Features:
- Associative Recall: Like the human mind, retrieving one memory triggers related ones through graph connections.
- Narrative Continuity: Automatically links events in temporal chains to understand "stories", not just facts.
- Psychological Profiling: Explicitly models "Psyche" (traits, values) separate from "Episodes" (events).
- Dynamic Knowledge Graph: Automatically builds a graph from unstructured text.
- Typed Memory System: distinct
Episode,Psyche, andGoalnodes. - Temporal Chaining: Narrative continuity for life-logging and long-term companions.
- Smart Retrieval: Hybrid RAG combining vector similarity with graph traversal.
- Structured Insights: Ask questions and get JSON data, not just text.
Currently, Persona supports Neo4j (via Docker) as the primary graph backend.
🚧 Coming Soon: We are actively working on decoupling the storage layer to support Qdrant, FalkorDB, and generic vector/graph stores.
git clone https://github.com/saxenauts/persona.git
cd personaCreate a .env file:
# Graph Database (Neo4j)
URI_NEO4J=bolt://localhost:7687
USER_NEO4J=neo4j
PASSWORD_NEO4J=your_secure_password
# AI Services (OpenAI - easiest setup)
LLM_SERVICE=openai/gpt-4o
EMBEDDING_SERVICE=openai/text-embedding-3-small
OPENAI_API_KEY=your_openai_api_keyStart the stack:
docker compose up -dExplore the API at http://localhost:8000/docs.
- Intelligent Forgetting: Decay mechanisms for irrelevant memories.
- Agentic Updates: Self-healing graph that corrects its own contradictions.
- Real-time Context: Streaming updates for low-latency personalization.
- Cross-User Intelligence: (Planned) Privacy-preserving shared insights.
(See internal TODO.md for the complete backlog)
MIT License. See LICENSE for details.