supermemory

supermemory builds an infrastructure layer for memory and continual learning that helps agents retain and improve over time. It extracts learning signals from the context of every user, task, and tenant, then stores what it learns in a vector-graph memory database.

To keep memory grounded and current, supermemory brings relevant knowledge back into the agent’s context in real time. It does this by injecting tokens during inference, so the agent can use the most relevant remembered information for the current run—while keeping learning and memory outside the model provider.

It also supports richer document-based learning. You can use document extraction with PDF handling that includes OCR, plus figure and diagram descriptions, enabling memory workflows that capture more than plain text. supermemory accepts multiple input types including files, chats, URLs, and raw data, giving you flexibility in where your learning signals come from.

You can deploy supermemory through an API, plugins, or MCP, with options intended for production and restricted environments such as air-gapped deployments. Deployment can be done in a data center, VPC, or on a laptop. Enterprise security and compliance information includes SOC 2 certification, HIPAA compliance, and GDPR alignment, along with contract signing and a DPA.

What is supermemory used for in agent systems?

supermemory extracts learning signals from user, task, and tenant context, stores them in a vector-graph memory database, and then injects relevant memory back into the agent’s context during inference.

How does supermemory keep memories up to date over time?

A learner component updates stored memory as new context arrives, merging, inferring, and forgetting against the memory database so the agent receives the most relevant information in subsequent runs.

Does supermemory work with any model or agent framework?

Yes. Learning and memory are designed to live outside model providers, so you can use them across models and harnesses while keeping the process in-context and interoperable.

What input types can I use to generate memory?

You can send files, chats, URLs, and raw data. For documents, supermemory supports PDF extraction with OCR and can describe figures and diagrams for richer memory signals.

How do I integrate supermemory into my system?

You can integrate via an API, plugins, or MCP. Use hooks to inject memory into your agent’s context.

Can supermemory be deployed in restricted or air-gapped environments?

Yes. It can run in air-gapped environments and can be deployed in a data center, VPC, or on a laptop. Enterprise security includes SOC 2 certification, HIPAA compliance, and GDPR alignment, with options for signing contracts and a DPA.