Isioma Anofienam

Full permissions, zero blast radius — the only agent policy engine that operates outside the LLM's reasoning path for verifiable security guarantees.

The policy layer for autonomous agents. Sits between your LLM agent and its tools, enforcing deterministic rules on every tool call. No LLM in the security path — fail-closed architecture. Works natively with Claude Code hooks, OpenCode plugins, and acts as an MCP proxy for Cursor, Windsurf, and any MCP-connected agent.

Available on npm as the @memstack suite — core, server, CLI, and MCP adapter — the neutral adapter layer above every memory provider.

Open-source TypeScript memory pipeline for AI agents. 21 storage adapters in a single package — from in-memory to pgvector to Mem0 — with a unified interface and token-budgeted context assembly. The ecosystem spans a headless server with HTTP API, a CLI for managing memory environments, and an MCP server that plugs agent memory directly into Claude Code, Cursor, and other MCP-compatible tools.

Shipped to TestFlight — the only baby tracker with embedded vector memory, similarity search, and a proactive AI co-pilot built in.

AI-native baby tracking app with a reasoning engine powered by Claude. Parents log feeds, sleep, nappies, and milestones in natural language — the backend converts every entry into vector embeddings and stores them in Cloudflare R2. A background pipeline performs vector similarity search across historical entries to assemble a token-budgeted context window, surfacing proactive insights like pattern detection, anomaly flags, and developmental guidance. Handles streaming responses from the LLM with chunked delivery for real-time UX. Currently in TestFlight — launching on the App Store soon.