Hexis Agent Project: Focuses on Memory, Identity and Emotion-based Decay

Hexis is an ongoing agent project explicitly distinguished by its author from tools like OpenClaw and Hermes Agent. While the latter focus on assistant capabilities, Hexis centers on memory and identity. The project aims to build agents with long-term memory and consistent personality, demonstrating this with the Samantha persona.

Key Design and Technical Details

Hexis is far from a lightweight demo; it is backed by a robust codebase with significant backend depth, featuring numerous migrations and SQL functions. The system's core logic resides within the database layer, while also offering UI and CLI interfaces for external interaction.

For its memory system implementation, the project combines AGE, pgvector, and relational tables to achieve joint retrieval across three types of data. Its standout design feature is human-like memory "decay": memories fade over time, but the rate of decay is dictated by emotional intensity. For instance, emotionally charged events (like a "first time") are forgotten much more slowly than ordinary ones.

2026-07-23 ~ 2026-07-23 · 5 related posts

• [source] Hexis shifts agent design from assistants to long-term memory and identity — QuixiAI · 2026-07-23
• Hexis leans into memory and identity, with a Samantha persona demo — QuixiAI · 2026-07-23
• [source] A memory project that looks more like an agent infrastructure codebase — QuixiAI · 2026-07-23
• [source] An AI memory system combines AGE, pgvector and relational tables with emotional decay — QuixiAI · 2026-07-23
• A lightweight memory system puts the logic in the database and adds UI and CLI — QuixiAI · 2026-07-23