MindCore

A standalone Rust crate providing a pluggable, feature-gated memory engine for AI agent applications.

Handles persistent storage, keyword search (FTS5), vector search (candle), hybrid retrieval (RRF), graph relationships, memory consolidation, cognitive decay modeling, and token-budget-aware context assembly.

Design Principles

• Library, not framework — projects call into MindCore, not the other way around
• Feature-gated everything — heavy dependencies behind compile-time flags
• Local-first — SQLite-backed, single-file databases, no cloud dependency
• Pure Rust where possible — candle over ort, SQLite over Postgres
• Proven patterns only — every component backed by research or established practice

Status

v0.1.0 published on crates.io. See:

• MINDCORE_ARCHITECTURE.md — full crate structure and API design
• MINDCORE_RESEARCH.md — research, landscape analysis, and specification
• DECISIONS.md — architectural decisions log
• research/ — competitive landscape analysis