Show HN: A memory database that forgets, consolidates, and detects contradiction
YantrikDB introduces a novel cognitive memory engine designed to prevent AI agents from getting "noisy" as their memory grows, unlike traditional vector databases. It intelligently forgets, consolidates information, and detects contradictions, moving beyond simple search to offer a more human-like memory experience for AI. This project stands out on HN for its deep technical approach to a critical AI scaling problem and its robust, production-ready engineering.
The Lowdown
YantrikDB presents itself as a solution to the limitations of current AI memory systems, particularly vector databases, which tend to degrade in recall quality after accumulating a large number of memories due to a lack of management features. This new engine aims to provide AI agents with a more sophisticated, cognitive memory capability by mimicking human-like memory functions.
- Core Problem Addressed: Traditional vector databases merely store memories, leading to recall degradation as the volume of data increases due to no consolidation, forgetting, or conflict resolution.
- Cognitive Memory Engine: YantrikDB actively manages memories through three primary mechanisms: forgetting (temporal decay with configurable half-life), consolidating duplicate memories, and detecting factual contradictions.
- Deployment Flexibility: It can be used as a standalone network server (Rust binary, Docker, Kubernetes), an MCP (Memory, Cognition, Perception) server for direct AI agent integration, or an embedded library for Python and Rust applications.
- Advanced Scoring & Management: Memories are scored using a multi-signal approach that considers semantic similarity, temporal decay, importance, graph proximity, and retrieval feedback. It also supports different memory types (semantic, episodic, procedural) with rich metadata.
- Architectural Robustness: The system features an embedded, local-first design with five distinct indexes (Vector, Graph, Temporal, Decay, Key-Value) and utilizes CRDTs for multi-device synchronization, ensuring conflict-free merging and data integrity.
- Performance and Stability: Despite its alpha status, YantrikDB boasts competitive performance metrics for recall and write operations and has undergone extensive hardening through chaos testing, fuzzing, and runtime deadlock detection.
- Distinction from Existing Solutions: The project highlights its advantages over vector databases (lacking decay/causality), knowledge graphs (poor for fuzzy memory), and memory frameworks (mere middleware), demonstrating significant token savings and improved precision in benchmarks.
By integrating advanced memory management principles, YantrikDB offers a compelling alternative to conventional AI memory solutions. Its focus on autonomous knowledge management—forgetting, consolidating, and resolving conflicts—aims to foster more coherent and reliable AI agent interactions, addressing a significant bottleneck in developing truly intelligent systems.