supermemory

What it does

supermemory takes in what an agent sees: conversations, uploaded files, URLs, and synced sources like Google Drive or Notion. Its learner model turns that material into facts, connects them to existing memories and time, and returns them as a user profile, a search result, or a virtual filesystem the agent can grep. Each input can become several atomic memories linked in a graph that updates, extends, derives and forgets as information changes.

People use it through a TypeScript or Python SDK, a REST API, a remote MCP server, or plugins for coding agents such as Claude Code and Cursor. Profiles provide standing facts attached to each turn without a search, and recall combines semantic search with graph context in one call. The product states that learning and memory are kept separate from model providers, so the same memory can serve different models and harnesses. Setup is an API key from the console; the local server runs from one binary.

It suits developers building agents that need persistent per-user or per-tenant memory, and enterprises needing SOC 2, HIPAA or self-hosted deployment. Hosted use requires an account and API key; the free plan includes $5 of credits renewed monthly. Limits to note: advanced PDF extraction with OCR is restricted to the Scale and Enterprise plans, and ingestion is asynchronous, so new memories must finish processing before they appear in search.

What it can do

• Extracts facts from conversations, files, URLs and synced sources into atomic memories
• Updates, extends, derives and forgets memories as information changes, with history kept for audit
• Maintains per-user or per-tenant profiles returned in a single call without a search
• Hybrid semantic search over memories and document chunks with graph edges and filters
• Mounts memory as a virtual filesystem that agents can ls, cat and grep
• Runs locally as a single binary or hosted in cloud, VPC or dedicated infrastructure

Works with

• Google Drive
• Gmail
• Notion
• OneDrive
• GitHub
• Granola
• Amazon S3
• Web crawler
• Claude Code
• Cursor

Before you choose

• Ingestion is asynchronous; new memories must finish processing before they appear in search or profiles
• Advanced PDF extraction with OCR is available only on Scale and Enterprise; Free, Pro and Max use standard extraction
• Usage is billed in SM tokens and per-query/operation meters; top-up credits do not expire but subscription credits reset monthly

Listed under

• Coding
• MCP and agent tools
• Agents
• AI agents
• AI memory
• API
• Context engineering
• Continual learning

Get started

Free plan starts from the billing page; API keys are created in the console.

1. Create an account at the console and get an API key from the billing/console page
2. Install the TypeScript or Python SDK and set SUPERMEMORY_API_KEY, or connect the remote MCP server at mcp.supermemory.ai/mcp using OAuth
3. Ingest content with add() and wait for processing before querying with search() or profile()

Subscription credits reset monthly; usage is metered in SM tokens and per query/operation at the published rate card. Free plan provides $5 of credits renewed each month, not a time-limited trial. Enterprise uses committed-spend pricing via sales.

Source: supermemory.ai/pricing, checked 29 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• supermemory — Memory and continual learning for agents ↗supermemory.ai
• Product — supermemory ↗supermemory.ai
• Introduction - supermemory | Memory API for the AI era ↗supermemory.ai
• Pricing — supermemory ↗supermemory.ai
• smfs - Supermemory File System ↗smfs.ai

Company website

supermemory.ai: product, pricing and policy pages

YC directory mirror

Batch and company name

Web search

Confirms the official site

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

manual_review

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.