Blog & Research

A retrospective on building ContextFS with AI assistance and how AI tendency toward MVP solutions creates predictable category errors that require human architectural intervention.

Enabling persistent, structured knowledge across AI tools and sessions through a formally-specified type system with 22 memory categories and sub-50ms query latency.

The traditional test pyramid was designed for human cognitive limitations. When AI generates entire features in single iterations, we need a new approach: Intent-Behavioral Testing (IBT).

A formal type-theoretic framework for AI memory systems, defining schema-indexed memory types (Mem[S]) and versioned memory with change tracking through a rigorous type grammar.

A comprehensive framework for AI-native workflows where developers describe intent rather than prescribe implementation, achieving 3-10x productivity gains with persistent memory systems.

A novel approach to multi-device LLM context management using vector clock algorithms traditionally used in distributed databases, enabling consistent AI memory synchronization across devices.

A structural approach to composable storage backends using protocol-based polymorphism with runtime capability detection, enabling seamless coordination of relational, vector, and graph databases.

A formal type-theoretic framework for understanding AI prompt engineering, drawing parallels between protein folding and context design to build more reliable AI systems.

Today we're launching ContextFS, an open-source memory layer that gives AI assistants the ability to remember context across sessions, tools, and devices.

A deep dive into how ContextFS uses ChromaDB and sentence-transformers to enable semantic search across your memories—finding relevant context even when you don't remember the exact words.

Learn how ContextFS implements AES-256-GCM encryption to ensure your memories remain private—even when syncing to the cloud. We can't read your data, and neither can anyone else.

Learn how to use ContextFS to orchestrate multi-step AI agent operations with persistent state, enabling complex workflows that span multiple sessions.