supermemory™

Purpose

supermemory is positioned as the "default memory and continuous learning engine" for AI agents. Its core goal is to free AI from relying on the model's own real-time learning capability, and instead provide personalization and continuous improvement for any model and any harness through an independent, interoperable memory layer. The official description points out that although pretraining and post-training can continue to improve model intelligence, they cannot enable models to learn and improve in real time, so memory needs to be decoupled from the model provider.

Main Capabilities

1. Universal memory API: provided in three ways: API, plugin, and MCP, and can be integrated with any model and harness.
2. Continuous learning model learner-1: able to extract and "dreaming" learning content from the context of any user, task, or tenant, forming dense, interconnected, continuously growing learning memory.
3. Vector graph database: stores learning results in a vector graph database, and brings them back to the model at runtime by injecting tokens into the model context in real time.
4. Infinite context and simplified retrieval: provides agents with infinite context and simplified retrieval capabilities.
5. Document extraction: Scale and Enterprise versions include advanced document extraction for PDFs, supporting OCR as well as descriptions of charts and diagrams; Free, Pro, and Max versions use standard PDF extraction.
6. Deployment flexibility: supports running in the user's data center, VPC, laptop locally, and even in air-gapped environments, with data able to remain entirely within the domain.
7. Security and compliance: SOC 2 certified, HIPAA compliant, GDPR aligned, and able to sign contracts and DPAs.

How to Use

• Integrate with existing agents or models through the officially provided API, plugin, and MCP.
• Connect the model to learner-1 so that it continuously learns from user, task, or tenant context, with the learning content written to the vector graph database.
• At runtime, the system injects memory into the model context in real time as tokens, so the model can gain personalized memory without retraining.
• Choose a data center, VPC, local, or air-gapped environment according to deployment needs.

Applicable Scenarios

• AI agents that need long-term memory and personalization.
• Teams that are already running memory in production and want to reduce token consumption and response time.
• Environments where multiple models and multiple harnesses coexist and require a memory layer that interoperates across models.
• Industries with strict requirements for data residency and compliance, such as healthcare (HIPAA) and EU (GDPR)-related businesses.
• Enterprises that need to deploy memory capabilities in air-gapped or local environments.

Performance Data

Official cited results include: average response time reduced from 40 seconds to 12 seconds, and token usage reduced by about 40%–50%.

Limitations

• Advanced PDF document extraction (including OCR and chart descriptions) is provided only in Scale and Enterprise versions; Free, Pro, and Max use standard PDF extraction.
• Regarding specific pricing, the complete list of feature differences across plans, API endpoint details, full integration documentation for plugins and MCP, and the list of supported languages and platforms, the current materials do not yet provide this.