Supermemory

Supermemory is a memory and continual learning layer for AI agents. It ingests files, chats and URLs, stores extracted learnings in a vector graph database, and injects the relevant ones into an agent's context at query time. It is offered through an API, MCP connections to ChatGPT and Claude, and plugins for Cursor, Claude Code and Hermes.

Supermemory is infrastructure for giving AI agents long term memory. Raw inputs such as files, chats, URLs and other raw data are fed into a model the company calls learner-1, which extracts, merges, infers and forgets information before writing it to a vector graph database. Relevant memories are then pulled back into an agent's context window in real time, so the agent can carry prior knowledge into a new task without retraining the underlying model. The design goal is portability. Memory sits outside the model providers, so a user's or tenant's accumulated context follows them across models and harnesses. The vendor positions this against simply stuffing documents into a context window, publishing internal runs that show token cost growing more slowly than markdown based memory as a corpus scales from a few hundred to nearly ten thousand files. Public benchmark claims cover LongMemEval, LoCoMo and ConvoMem, and an independent SWE-ContextBench evaluation is cited for software task resolution. In practice the product reaches users in three ways: a direct API for teams building their own agents, MCP based connections to assistants such as ChatGPT and Claude, and plugins for coding tools like Cursor, Claude Code and Hermes. A console handles account and project setup. The company says the system serves over 100k organizations and processes more than a trillion tokens per month, with recall latencies in the low hundreds of milliseconds per query. Supermemory competes with agent memory libraries such as Mem0, and its differentiators are the graph based storage, a time aware forgetting mechanism, self hosted deployment, and compliance coverage. Teams with strict data residency needs can run it in their own data center, inside a VPC, or air gapped. Free, Pro and Max tiers ship standard PDF extraction, while Scale and Enterprise add OCR plus descriptions of figures and diagrams. Whoever picks it up should expect to wire an API or MCP endpoint rather than click through a finished consumer app.