Skip to content

Our Story

Solo founder. Bootstrapped. Building in public.

I'm Blake Heron, and I built Novyx because I kept running into the same problem: AI agents forget everything.

I was building agents for side projects — personal assistants, research tools, workflow automations. Every single one had the same flaw. You'd spend 30 minutes teaching it your preferences, your project context, your constraints. Then the session ends. Next time? Blank slate. All that context, gone.

I looked for solutions. LangChain has memory classes, but they're in-process — restart your app and it's gone. Vector databases can store embeddings, but they don't understand memory lifecycle. There's no rollback, no audit trail, no way to undo a bad write. You're duct-taping infrastructure together and hoping it holds.

So I built what I wanted: a memory API that just works. Store a memory, recall it semantically, roll it back if something goes wrong. Three lines of code.

That became Novyx Core.

What Novyx actually is

Novyx Core is a memory-as-a-service API for AI agents. You call remember(), recall(), and rollback(). Under the hood, every write is cryptographically hashed, every operation gets an audit trail, and semantic search finds relevant memories in milliseconds.

I built it on Supabase and Postgres because I wanted something reliable, not clever. The API runs on Fly.io. There's a Python SDK, a JavaScript SDK, integrations for LangChain, CrewAI, LlamaIndex, OpenClaw, Cursor, and Claude Code. An MCP server with 107 tools. Everything is real, everything is live, and everything is documented.

To prove the API works, I built Novyx Vault — an open-source AI notes app built entirely on Novyx Core. It's not a demo. It's a real app that uses every feature: persistent memory, knowledge graph, rollback, semantic search. If the API can power Vault, it can power your agent.

3

lines to integrate

107

MCP tools

185+

API endpoints

Where this is going

This is a solo operation, bootstrapped, no VC. I ship fast and I ship in public. The roadmap is driven by what real users need, not what looks good in a pitch deck.

I'm working on multi-agent memory sharing (Spaces), autonomous memory maintenance (Cortex), and time-travel debugging (Replay). SOC 2 compliance is on the roadmap. The free tier is generous because I want people to actually use this and tell me what's broken.

If you're building AI agents and you're tired of the memory problem, try Novyx. If something doesn't work, tell me. I read every message.

— Blake Heron@BlakeHer_on