Skip to main content
StudioMeyer

3D VISUALISATION · 111 ENTITIES · 140 RELATIONS

What memory actually looks like.

Three months of project work as a 3D knowledge graph. Real data from real sessions, nothing made up. This is what AI memory looks like when an entire project lives inside: every decision, every contributor, every connection.

Drag · scroll to zoom · click nodes for detail

entities
111
relations
140
AI agents
7
Deutschland
EU

What you're looking at

One story, three chapters.

Every node a real step. Every connection a real decision. Nothing generated. Three months of project work from February to April 2026, one to one, the way memory recorded it.

FEBRUARY 2026

Concept emerges

An idea at the start: a small learning platform around AI tools. The first notes pile up, tech stack, audience, name. Small, messy, full of open questions. Memory keeps every step, even the dropped ones.

  • Memory-First curriculum
  • Next.js + MCP stack
  • Brand: "Academy"

MARCH 2026

Agent fleet grows

Seven AI agents join, each specialised on one job. First learning content takes shape, connections cross-link in every direction. Memory sees: things are growing together.

  • 7 agents in the fleet
  • Skill pattern: success-rate-pill
  • Recipe layer Phase B

APRIL 2026

Course correction

Late insight: trying to sell too early slows the learning. Course correction, everything free, focus on what works. Notes turn into a visible knowledge network. Memory shows the turning point right in the graph.

  • MONETIZATION_VISIBLE = false
  • Phase 2+3 visualisation LIVE
  • 3D force graph demo

What memory actually does

Not "remember user preferences".

Memory stores what matters, connects it in the knowledge graph, forgets what gets stale, and delivers exactly the context the model would otherwise have to invent.

Knowledge graph with evidence

Entities plus typed relations with strength and evidence. Multi-hop traversal in a single tool call. You see above how projects, people, MCP servers and infrastructure connect to each other.

Hybrid search

Trigram plus full-text plus semantic embeddings plus RRF reranking. Finds "the SSL thing" even if you stored it as "Certbot renewal". Question-conditioned retrieval distinguishes episodic vs. semantic queries.

Confidence plus decay

Not all entries are equal. Confidence per learning, bi-temporal valid_from/valid_to, contradiction detection, automatic decay for stale memory. Plus point-in-time queries via asOf param.

MCP-native

56 tools that Claude, Cursor, Codex and Claude Desktop understand instantly. OAuth 2.1 with PKCE plus magic-link email plus Google, GitHub and Discord social login. No setup beyond a bearer token.

Multi-tenant in the EU

Supabase Pro Frankfurt with row-level security per tenant, active-standby HA with Hetzner floating IP, daily backups plus restore tests. GDPR compliant, IPv4 plus IPv6.

Import plus time travel

Import ChatGPT, Claude and Gemini history. Session start auto-loads sprint, recent sessions, top learnings and personality. The replay slider walks through every step of a past session as a timeline.

Connect in 30 seconds

One URL. No password. No setup.

Memory runs as a hosted MCP server. No server of your own, no database, no API keys in your .env. Just one URL and a login.

  1. In Claude Desktop: Settings → Connectors → Add custom connector
  2. Paste URL: https://memory.studiomeyer.io/mcp
  3. Authorise via email (magic link) or Google/GitHub/Discord
  4. Claude Code: claude mcp add studiomeyer-memory --transport http https://memory.studiomeyer.io/mcp

Health check and OAuth discovery are publicly reachable.

Ready for your own memory?

Free plan forever, Pro from 19 EUR/month. One URL and your AI never forgets, no matter which tool you use.

See Memory