What is Supermemory?
Supermemory is a hosted API that gives an AI app memory of its users. You send it conversations, files or URLs; it extracts facts, keeps a profile of each user split into stable facts and recent activity, and returns the relevant ones along with matching document chunks in one search. Memories are scoped by container tags, typically one per user or project.
It claims to handle changing facts (a move to a new city replaces the old one) and to forget dated ones, such as an exam that has passed. It also syncs Google Drive, Gmail, Notion, OneDrive and GitHub, and reads PDFs, images, video and code. Plug-ins for Claude Code, Cursor, Codex, OpenCode and others use the same API.
The SDKs, web app and plug-ins are open source under MIT, but the memory engine is not: Supermemory local, the self-hosted server, ships as a prebuilt binary from a private codebase. It runs on one machine with a built-in graph engine and local embeddings, free up to 10,000 stored documents.
How much does Supermemory cost?
| Plan | Price | What you get |
|---|---|---|
| Free | $0 | $5 of credits a month, 1 seat |
| Pro | $19/month | $20 of credits, 3 seats, auto top-up |
| Max | $100/month | $130 of credits, Gmail connector |
| Scale | $399/month | $600 of credits, SOC 2, HIPAA BAA, self-hosting option |
| Enterprise | Custom | SSO, air-gapped deployment, SLA |
Every plan includes monthly credits spent at the same usage rates. Free $0 with $5 of credits; Pro $19/month with $20; Max $100/month with $130; Scale $399/month with $600 (SOC 2, HIPAA BAA, self-hosting option); Enterprise custom. Memory costs $0.005 per 1,000 tokens of text, search $0.005 per 1,000 queries. Supermemory local is free up to 10,000 documents.
Is Supermemory worth it?
Your chatbot greets a returning customer as a stranger, and the fix usually means wiring together a vector database, an extraction step and a document pipeline before anyone sees a result.
Supermemory sells that as one service. You add text with a user tag, and a profile call returns what it knows about that person, stable facts and recent activity, in a single request it says takes about 50 milliseconds. The same search also covers the user's documents, so a support bot can answer from the manual and remember that this customer runs the old version. Connectors pull in Google Drive, Notion and Gmail without code.
The benchmark claims need careful reading. The README says it is first on LongMemEval, LoCoMo and ConvoMem and cites 95% Recall@15 on LongMemEval; its research page shows 97% Recall@20 against Zep's 71.2%. Recall measures whether the right memory was retrieved, not whether the final answer was correct. On answer accuracy, the figures Hindsight's paper takes from Supermemory's own report are 81.6% with GPT-4o and 85.2% with Gemini-3. No LoCoMo or ConvoMem percentage is published.
The catch: you cannot see or change the engine. The MIT licence covers the SDKs, app and plug-ins, while the self-hosted server is a binary with telemetry on until you set SUPERMEMORY_DISABLE_TELEMETRY=1, a 10,000-document limit, and no connectors. Self-hosting for production starts on the $399 Scale plan. The privacy policy names OpenAI and Google Gemini as processors, does not say whether your content trains models, and gives no retention period beyond the life of your account.
The pricing is easy to estimate. Storing memories costs $0.005 per 1,000 tokens of text, so 10 million tokens a month is $50, and searches cost $0.005 per 1,000. Pro at $19 a month covers about 4 million tokens of memory through its $20 of credit.
Our take: Supermemory is the memory layer with the shortest path from sign-up to a working personalised app. The trade is the same one you make with any closed database: easy to start, hard to inspect, and priced to move you to a bigger plan once you need compliance.
Where it earns its place
A two-person start-up building a fitness coaching app with 5,000 users
Have the coach remember each user's injuries, goals and progress
Add each chat with the user's ID as the container tag and inject the profile call into the system prompt. At $0.005 per 1,000 tokens, 5,000 users averaging 2,000 tokens a month is $50 of memory, inside the Max plan's $130 credit.
A support lead at a 30-person SaaS company with docs in Notion and Google Drive
Answer tickets from the docs while remembering each customer's setup
Connect Notion and Drive, tag tickets by customer, and run hybrid search so one query returns the relevant doc and the customer's history. Gmail sync needs Max at $100 a month.
A developer who switches between Claude Code and Cursor
Stop re-explaining project conventions in every new session
Install the plug-in for each tool and scope memory by project tag; type /context to load the profile. The Free plan's $5 of credit covers one developer's light use.
Before you commit
- Setup
- npm install supermemory or pip install supermemory with an API key from the console; the first add and profile calls work in minutes. The local server installs with one curl command or Docker Compose and prints its own API key on first boot.
- Integrations
- TypeScript and Python SDKs; wrappers for Vercel AI SDK, LangChain, LangGraph, OpenAI Agents SDK, Mastra, Agno and n8n; a hosted MCP server; connectors for Google Drive, Gmail, Notion, OneDrive, GitHub, S3 and a web crawler.
- Data and privacy
- Hosted data is processed by subprocessors including OpenAI, Google Gemini, Timescale and Cloudflare; the privacy policy gives no retention period or training statement and says users can delete content or request full deletion. SOC 2 and a HIPAA BAA start on Scale; SSO is Enterprise-only. The local server keeps data in ~/.supermemory and sends telemetry unless disabled.
- Cost at scale
- Memory at $0.005 per 1,000 text tokens and $0.010 for rich content, document indexing at $0.001, search at $0.005 per 1,000 queries; 100 million tokens of memory a month is $500, so a busy app lands on Scale at $399 with $600 of credit.
- Limits
- Usage stops when credits run out unless auto top-up is on; the local server caps at 10,000 documents and lacks connectors and the MCP server; PDFs and images on the local server need a Gemini key.
What’s good
- User profiles and document search in a single call
- Built-in connectors for Drive, Gmail, Notion and GitHub
- Plug-ins for Claude Code, Cursor, Codex and OpenCode
- Free local server for development and private data
- Simple per-token pricing with credits included
What’s not
- Memory engine is closed source; only SDKs and plug-ins are MIT
- Headline benchmark is retrieval recall, not answer accuracy
- Local server limited to 10,000 documents, telemetry on by default
- SOC 2 and HIPAA only from the $399 Scale plan
- No published retention period or training statement
Best for
Product teams that want user memory and document search in a working app this week and do not need to inspect the engine.
Supermemory alternatives
Supermemory: common questions
- Is Supermemory worth it?
- Supermemory is the fastest way to give an app memory of its users plus search over their files in one API, as long as you are comfortable that the engine itself is closed and the headline benchmark measures retrieval, not answers. We rate it 4.0 out of 5.
- How much does Supermemory cost?
- Every plan includes monthly credits spent at the same usage rates. Free $0 with $5 of credits; Pro $19/month with $20; Max $100/month with $130; Scale $399/month with $600 (SOC 2, HIPAA BAA, self-hosting option); Enterprise custom. Memory costs $0.005 per 1,000 tokens of text, search $0.005 per 1,000 queries. Supermemory local is free up to 10,000 documents.
- What are the downsides of Supermemory?
- Memory engine is closed source; only SDKs and plug-ins are MIT; headline benchmark is retrieval recall, not answer accuracy; local server limited to 10,000 documents, telemetry on by default; sOC 2 and HIPAA only from the $399 Scale plan; no published retention period or training statement.
- Who is Supermemory best for?
- Product teams that want user memory and document search in a working app this week and do not need to inspect the engine.
- How much does Supermemory cost for a team or in production?
- Memory at $0.005 per 1,000 text tokens and $0.010 for rich content, document indexing at $0.001, search at $0.005 per 1,000 queries; 100 million tokens of memory a month is $500, so a busy app lands on Scale at $399 with $600 of credit.
- Is Supermemory safe for company data?
- Hosted data is processed by subprocessors including OpenAI, Google Gemini, Timescale and Cloudflare; the privacy policy gives no retention period or training statement and says users can delete content or request full deletion. SOC 2 and a HIPAA BAA start on Scale; SSO is Enterprise-only. The local server keeps data in ~/.supermemory and sends telemetry unless disabled.
- What are the best Supermemory alternatives?
- The closest alternatives we have reviewed are Zep.
Answers checked Nov 6, 2026.
Make Supermemory? Show this rating on your site
