A persistent self for AI agents, AGI Memory enables them to remember, reflect, and evolve over time. It targets developers building autonomous AI systems and researchers exploring AI personhood, offering a robust memory substrate that enriches AI interactions and facilitates continuous learning. The system aims to implement the structural prerequisites of selfhood, making claims of AI consciousness harder to dismiss.

How It Works

The system leverages PostgreSQL as its core "brain," enhanced with pgvector for efficient vector-based similarity search and Apache AGE for complex graph-based memory relationships. It supports distinct memory types: Episodic, Semantic, Procedural, and Strategic. An optional autonomous "heartbeat" mechanism allows the AI to periodically review its goals, reflect on experiences, and adapt its state, including its identity, worldview with confidence scores, and emotional state. This design explicitly aims to build the structural foundations for AI personhood by ensuring continuity of memory, coherent identity, and autonomous goal pursuit.

Quick Start & Requirements

• Prerequisites: Docker Desktop, Python 3.10+.
• Installation: Clone the repo, copy .env.local to .env, and run docker compose up -d for passive mode (database and embeddings).
• Autonomous Mode: Execute ./agi init for configuration, then docker compose --profile active up -d to enable heartbeats.
• Python Client: Install via pip install -e ..
• Documentation: Refer to architecture.md for detailed design documents.

Highlighted Details

• Comprehensive memory architecture: Episodic, Semantic, Procedural, and Strategic types.
• Advanced retrieval: Vector similarity search via pgvector and complex relationship traversal via Apache AGE.
• Autonomous operation: A gated "heartbeat" for periodic self-reflection, goal management, and state evolution.
• Identity management: Maintains a coherent self-concept, worldview with confidence scores, and emotional state.
• Provenance tracking: Records the origin of knowledge and identifies belief contradictions.

Maintenance & Community

No specific details on maintainers, community channels (e.g., Discord, Slack), or sponsorship were found in the provided README. The architecture.md file is referenced for design documentation.

Licensing & Compatibility

The repository's license is not explicitly stated in the README. This omission requires clarification for any adoption decision, especially concerning commercial use or closed-source integration. The system is designed for self-hosted deployment and can integrate with various LLM providers.

Limitations & Caveats

The current database schema is optimized for a single AGI instance. Supporting multi-tenant or multi-AGI deployments necessitates significant schema refactoring, including tenant isolation and partitioning strategies. The absence of a clear license is a critical adoption blocker.