SLIDE 1 / 5 — HOOK

FREE + OPEN SOURCE

15 Memory Tools.
Zero Enforcement
Tools.

Until now.

SLIDE 2 / 5 — THE LANDSCAPE

Every MCP memory tool does the same thing.

claude-mem 37,200 ⭐ — memory
Mem0 OpenMemory VC-funded — memory
Official Anthropic server 61K downloads — memory
15+ others store → recall — memory
They all require the agent
to cooperate.

SLIDE 3 / 5 — THE PROBLEM

What happens when Claude
doesn't cooperate?

❌ Context gets compacted → rules forgotten
❌ Agent doesn't call recall() → memory unused
❌ New session → blank slate, same mistakes
❌ CLAUDE.md → ignored under context pressure
Memory says "please remember."
Nothing stops the bad action.

SLIDE 4 / 5 — THE SOLUTION

Pre-Action Checks
don't ask. They enforce.

1 1st failure → feedback captured
2 2nd failure → pattern noted
3 3rd failure → prevention rule auto-generated
5 5th failure → BLOCKING GATE 🛑
No agent cooperation. Survives compaction. Hooks fire regardless.

SLIDE 5 / 5 — CTA

The only MCP tool that
learns AND enforces.

Memory + Enforcement > Memory alone

npx thumbgate init
GitHub: github.com/IgorGanapolsky/thumbgate
FREE + MIT LICENSED
Claude Code Codex Gemini Cursor Amp