Show HN: Shared memory graph for Claude and ChatGPT, over MCP

Hacker News Top Products

Summary

UML (Universal Memory Layer) is an open-source shared memory graph for Claude, ChatGPT, and other AI agents, letting users save and recall structured memory across tools via MCP with a visual dashboard and API.

No content available
Original Article
View Cached Full Text

Cached at: 07/31/26, 05:01 PM

# UML - Universal Memory Layer Source: [https://uml.gpmai.workers.dev/](https://uml.gpmai.workers.dev/) [![UML logo](https://uml.gpmai.workers.dev/assets/uml-icon.png?v=2)**UML**Universal Memory Layer](https://uml.gpmai.workers.dev/)![](https://uml.gpmai.workers.dev/assets/uml-icon.png?v=2)Universal Memory Layer ## Tell it once\. Every AI remembers\. UML is one private memory graph shared by Claude, your agents, and your apps — you can see it, edit it, and own it\. Free early access · Open source · You own your memory · Works with API, SDK, dashboard, and MCP\-linked clients\. Your memory, as a living graph — real product footage\. How it works ## Four steps\. Simple enough to explain to anyone\. 1 ### You talk Tell your AI:*"save this to uml — I finished Unit 3\."*Or let your app send it automatically\. 2 ### UML organizes it AI extracts the durable facts\. The backend verifies them and writes real structure — never a raw chat dump\. 3 ### You can see it Open the dashboard: a living graph, timelines, organized pages — and a receipt for every single save\. 4 ### Every tool remembers Claude, Cursor, your own agent — all connect to the same private memory\. Save once, recall everywhere\. Connect in 2 minutes ## Pick your tool\. Paste one link\. Done\. Sign up, create your private memory link in the Connect tab, then paste it into any MCP\-capable client — or call the API from your own app\. *uml · connect* **1**Sign up free → open*Connect*→*Create MCP link* **2**Claude → Settings → Connectors →*Add custom connector* **3**Paste your private link and start chatting: Custom connectors need a paid Claude plan\. Then just say:*"save this to uml"*or*"what do you know about me?"* **1**Create your MCP link in the Connect tab, then run: Claude Code gets three tools:*save\_memory*,*save\_conversation*,*recall\_memory*\. **1**Add UML to your MCP config \(*\.cursor/mcp\.json*\): Works the same in any MCP\-capable editor or agent host\. **1**One call: recall relevant memory*and*auto\-save the turn: Plus*/v1/save*,*/v1/recall*,*/v1/graph*,*/v1/rules*,*/v1/export*— same private graph\. Why a graph ## Structured memory, not transcript replay\. UML turns useful context from chats, events, documents, tools, and workflows into structured memory objects — entities, facts, events, relationships, pages, and receipts\. ### Chat history is not memory\. Messages are source material\. UML keeps the durable meaning that should survive beyond a single exchange\. ### Memory is structured meaning\. UML stores verified entities, facts, events, relationships, rollups, and receipts so systems can reason over them later\. ### Backend is the authority, not the LLM\. The LLM proposes; UML verifies; UML writes\. Recall and extraction stay separate\. What makes it different ## A graph that can update, not just append\. New evidence can create, update, promote, invalidate, or supersede memory\. UML is not an append\-only chat log\. ### A memory you can see Your facts live in a visual graph with clusters you can open, search, drag, and edit — not a hidden blob\. ### Latest truth wins Say "actually, that changed" — the old fact is superseded and kept as visible history on a timeline, not deleted, not duplicated\. ### Receipts for every save Each write returns proof: what was saved, updated, or refused — with source and timestamp\. No silent writes, ever\. ### Your rules Decide what gets collected and what must never be saved — plain\-language rules the backend actually enforces\. One graph, many inputs ## Chat is one input\. Not the whole product\. UML can ingest useful context from tools, apps, agents, docs, events, workflows, and custom schemas\. Chat turnsManual savesDocumentsTool outputsApp eventsSupport ticketsMeeting notesProject rulesWorkflow stateCustom schemas Private by default ## Your memory belongs to you\. Provably\. No ads\. No trackers\. No selling data\. No training on your memory\. Export everything as JSON or delete it all — any time, one click\. ### Privacy Policy UML does not sell user data or use user memory for unrelated purposes\. User memory belongs to the account that created it\. Connected tools are third\-party services and access UML only through private links or tokens you create\. ### Terms of Service UML is an early product and may change\. Use it responsibly\. You are responsible for tokens and private links, and you can revoke them from Connect\. Avoid sensitive or regulated data unless UML is appropriate for that use\. ## Build memory once\. Use it everywhere work continues\. Free early access — founding users keep the Pro tier free when paid plans launch\. ## Support Need help connecting UML or shaping an integration?

Similar Articles

rohitg00/agentmemory

GitHub Trending (daily)

agentmemory is an open-source persistent memory layer for AI coding agents (Claude Code, Cursor, Gemini CLI, Codex CLI, etc.) that uses knowledge graphs, confidence scoring, and hybrid search to give agents long-term memory across sessions via MCP, hooks, or REST API. Built on the iii engine, it requires no external databases and exposes 51 MCP tools.