Supermemory

Supermemory is context infrastructure for AI agents, providing long-term and short-term memory, retrieval, and personalization capabilities. It is designed to give agents a durable understanding of people and entities over time, rather than just acting as a search box over raw logs.

Core capabilities include:

• Memory & Continual Learning: A graph-based approach where facts update, connect, and forget in real time.
• SuperRAG (Retrieval): Grounding answers in documents and knowledge bases.
• User Profiles: Built-in static and dynamic context for personalization.
• Connectors & Extractors: Multimodal support for text, chats, PDFs, images, video, and code.
• Unified Context: Memory, profiles, and SuperRAG all share the same context pool when using the same isolation identifier.

Supermemory provides different levels of compliance depending on your plan.

• SOC 2 Type II: Certified. Available on production plans (typically Scale and above).
• GDPR: Compliant. Supports EU personal data handling, including access and erasure workflows.
• HIPAA: BAA (Business Associate Agreement) is available for eligible cloud plans (Scale / Enterprise).

To request a report, DPA, or BAA, contact support@supermemory.com or your enterprise contact.

The @supermemory/tools package provides specialized memory tools for three primary AI frameworks through dedicated submodule exports. Each submodule uses a function-based architecture optimized for its target framework:

• AI SDK: Tools compatible with the Vercel AI SDK.
• OpenAI: Tools compatible with the OpenAI SDK, including support for OpenAI Function Calling.
• Mastra: Tools compatible with the Mastra framework.

Refer to the specific submodule documentation for implementation details for each framework.

SMFS (Supermemory Filesystem) mounts your Supermemory container as a real directory, allowing AI agents to interact with memory using standard shell commands like ls, cat, and grep.

Instead of a traditional filesystem that requires manual tree walking or large file reads, SMFS provides:

• Semantic grep by default: Searching surfaces relevant information across the container ranked by meaning. Standard flags fall back to exact-match grep.
• Distilled Memory Paths: Files marked as memory paths are indexed by Supermemory to prevent context bloat.
• Virtual profile.md: A live digest located at the mount root. Agents can cat profile.md to get a one-shot summary of the container contents.
• Bidirectional Sync: Background synchronization ensures local reads hit a cache while writes are pushed to Supermemory.

It is important to distinguish between Retrieval-Augmented Generation (RAG) and Memory:

• RAG: Retrieves document chunks. It is stateless and returns the same results for everyone based on the provided context.
• Memory: Extracts and tracks facts about users over time. It understands temporal changes (e.g., knowing that "I just moved to SF" supersedes "I live in NYC").

Supermemory combines both by default, providing both knowledge base retrieval (RAG) and personalized context (Memory) in every query. It also features automatic forgetting, where temporary facts expire and contradictions are resolved automatically to prevent noise from becoming permanent memory.

Company Brain uses tool connectors (e.g., GitHub, Linear) to perform actions on your behalf. The behavior differs depending on whether the operation is a read or a write:

• Read from a tool: When you ask for information (e.g., "what are my open PRs?"), Company Brain attempts to use your personal connection first. If no personal connection is found, it falls back to the organization-shared connection.
• Write to a tool: When you command an action (e.g., "create a Linear issue"), the action always runs under your own account. This ensures that all changes, such as creating issues or comments, are correctly attributed to you in the target tool rather than appearing as a generic organization account.

If neither you nor the organization has the tool connected, Company Brain can request a teammate to lease temporary access for a single request.

Supermemory can be utilized in several ways depending on your needs:

• Developer Platform: Use the API, SDKs, and architectural concepts to build custom memory implementations for your own applications.
• Company Brain: Deploy a shared knowledge engine for teams to centralize and access collective intelligence.
• Plugins and MCP: Integrate memory directly into existing AI interfaces like Claude or specialized coding agents using the Model Context Protocol (MCP).
• Self-hosting: Run the entire infrastructure locally with zero configuration for privacy or development purposes.

Supermemory manages the lifecycle of information through:

• Automatic Forgetting: Temporary facts (e.g., "I have an exam tomorrow") automatically expire after the relevant date.
• Conflict Resolution: When new information contradicts old information, the system automatically resolves the conflict to ensure the user profile remains accurate.
• Noise Reduction: Prevents transient information from becoming permanent memory.