3D VISUALISATION · 111 ENTITIES · 140 RELATIONS
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
What you're looking at
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
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.
MARCH 2026
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.
APRIL 2026
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.
What memory actually does
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.
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.
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.
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.
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.
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 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
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.
https://memory.studiomeyer.io/mcpHealth check and OAuth discovery are publicly reachable.
Free plan forever, Pro from 19 EUR/month. One URL and your AI never forgets, no matter which tool you use.
See Memory