Supermemory is opening a memory graph in its developer console for teams adding persistent memory to AI agents. Chats, files, and URLs go in as documents and are broken into atomic facts, while a background process called dreaming keeps extracting those facts, linking related memories, and resolving updates. When a fact changes, search uses the latest version and the previous one stays in the graph as history. Builders can now inspect which facts were kept, linked, or superseded, which is hard to do when the only fallback is stuffing older transcripts back into the prompt. New facts connect in three ways. An update replaces what used to be true. An extend adds detail without invalidating the earlier memory. A derive is an inference drawn from patterns across memories that were never stated in one place. Short-lived facts can expire, and extraction and linking run on a specialized model, learner-1, so the agent's main model is not left to remember the full history itself. The console view is at console.supermemory.ai/graph, and the same visualization is available as the React package @supermemory/memory-graph. Supermemory was founded by Dhravya Shah and is based in San Francisco. The graph is reachable through the company's API, plugins, and MCP server, so stored memories can be searched or injected back into context across models and agent harnesses.