← Back
Early AI continuity prototype

MindHub

Memory and identity that persist across AI tools.

CLIsurfaceWebsurfaceCore intelligenceheuristic core + optional LLM augmentationIdentitypersistent assistantMemorythreads · context · retrievalPermissiongovernance · telemetryLLM providersOpenAI-compatible · Anthropic · local

The problem

AI assistants forget. Move between models, tools, or devices and you lose memory, identity, permissions, and project context. MindHub starts from a different premise: the assistant should live in a memory and identity layer, not inside any single model.

What exists today

An early working prototype. CLI and web surfaces; a persistent assistant identity; thread memory and context weaving; retrieval endpoints; support for multiple LLM providers (OpenAI-compatible, Anthropic, and local); and governance telemetry, documented across phased build notes.

What I directed, built, and tested

I architected the workspace — a core intelligence package, web and CLI surfaces, and shared libraries — and directed an approach that pairs a heuristic core with optional LLM augmentation. Through AI-assisted development I built identity persistence, thread and memory storage, retrieval, and governance instrumentation.

What I learned

Continuity is an architecture problem, not a prompt. Separating identity, memory, and permission layers is what lets an assistant move without losing itself — and local-first memory keeps it private and portable.

Next prototype step

Semantic and vector retrieval, cross-device sync, rehydration packets for fast context transfer, and an MCP server so any tool can read and write the same memory.