Teach your agent
to think better.
A versioned knowledge base that learns from its mistakes.
Introducing Helix 1.0
A recursive agent that learns from corrections to improve your knowledge base automatically.
Girl with a Pearl Earring — Johannes Vermeer, c. 1665
How it works
Three primitives
Everything you need to close the loop between agent output and knowledge quality.
Versioned documents
Markdown in Postgres with full version history. Every edit tracked. No deploy cycle — edits are live instantly.
MCP server
Agents connect via stdio. Six tools: read, search, list, import, report usage, propose corrections.
Feedback loop
Agent output gets corrected. The correction traces back to the source document. Knowledge updates. The agent improves.
Introducing
Helix
A recursive agent that analyzes human corrections in real-time, traces them to the source knowledge, and proposes targeted rewrites. It learns from its own mistakes — when you deny a recommendation, that feedback improves its future analysis.
Human corrects agent output
↓
Helix: "In messaging/tone-guidelines, change this section"
↓
You: Approve → knowledge updates instantly
You: Deny → Helix learns why it was wrong
Integrations
Works with your stack
Any framework that speaks MCP.
Without CortexKB
With CortexKB
Knowledge that improves itself.
Set up in minutes. Import your existing files. Connect your agent.
Get started freeCortexKB
Knowledge is the compound interest of intelligence.