Solitaire – identity layer for AI agents, not just another memory tool

Gallery

About

I built Solitaire because "memory-enabled" AI sessions still felt like starting over, just with slightly more context.The agent could retrieve facts from last week, but the collaboration itself never improved. An analogy I like is that it’s a “smart stranger with a better notebook.”The problem isn't retrieval. There are lots of tools that store and retrieve, to varying degrees of success. The problem is that nothing changes about how the agent works with you. It recalls what you said without learning how you think.Solitaire is my attempt at the next layer. It's identity infrastructure for AI agents, and is designed to make the interaction itself improve over time.What it does beyond memory: - Behavioral genome: disposition traits (observance, assertiveness, warmth, etc.) that evolve from real interaction, not static config. - Experiential memory: encodes how sessions felt, not just what was said. - Autonomous self-improvement: retrieval weights adjust based on what proved useful. The knowledge graph self-heals (contradiction detection, confidence rescoring, entity relinking). - Anticipatory retrieval: predicts what context you'll need and preloads it before you ask. - Guided onboarding: new users build a partner through conversation, not a JSON file. - Memory-compatibility: use Solitaire’s built-in memory or bring your own existing memory tools. - Model-compatibility: many work out of the box, we’re building towards integrating others. - Your data stays local, so you have full data sovereignty.600+ sessions, 14,000+ accumulated entries in real production use and counting. I have two research papers coming out of the longitudinal work.Repo: https://github.com/PRDicta/Solitaire-for-AgentsI know the agent memory space is popular, but memory is one component. Identity is the thing nobody's building. The agent should get better at working with you, not just better at remembering.Would especially welcome feedback on: 1. Does the identity vs. memory distinction make sense to you? 2. Where do current memory tools break down for you in practice? 3. What would make something like this worth integrating into your agent stack? 4. What am I missing?