Supermemory: The Memory Engine for AI Agents, Explained
Supermemory is a memory engine for AI agents. Developers send it conversations, documents and files, and it pulls out the facts worth keeping, stores them in a graph where facts update, connect and expire over time, keeps a profile for each user, and feeds the right context back to the AI model when it’s needed. It’s available as an API, as plugins for tools like Claude Code and Cursor, and as an MCP server you can add to ChatGPT or Claude. The company says it serves more than 100,000 organizations and over a trillion tokens a month, and its open-source repository has about 31,600 stars on GitHub.
The quick answer: Supermemory is one of the best-known names in AI memory, built by a young founder with heavyweight backers. It’s infrastructure for developers who want their apps and agents to remember users, and its consumer app has been retired in favor of the engine. A disclosure: we make Carly, an AI assistant with its own memory and workspace, so we compete on the memory side. Everything below comes from Supermemory’s site, docs and funding coverage.
Supermemory at a glance
Checked against Supermemory’s site, documentation, pricing and funding coverage on November 12, 2026.
| Supermemory | |
|---|---|
| What it is | A memory and context engine for AI agents and apps |
| How you use it | REST API, TypeScript and Python SDKs, coding-agent plugins and an MCP server |
| Works with | Claude Code, Cursor, Codex, ChatGPT, Claude, Grok, LangChain, Vercel AI SDK and more |
| Hosting | Supermemory’s cloud, a dedicated instance, self-hosted or a local binary |
| Open source | MIT-licensed repository with about 31,600 GitHub stars |
| Compliance | SOC 2 Type II, GDPR, HIPAA BAA on Scale and Enterprise |
| Company | Founded by Dhravya Shah in San Francisco; seed round led by Susa Ventures |
What Supermemory does
Supermemory’s homepage describes it as “building the default engine for memory and continual learning for agents.” The core argument is that AI models are already smart, but they don’t learn from each conversation. Supermemory’s answer is to keep memory outside the model, so it carries across whichever model or tool you use: “Learning and memory live outside the model providers, so they carry across models and harnesses and stay interoperable, in context, and yours.”
Its engine does four main things, according to its developer summary:
- A memory graph. Facts “update, extend and derive from each other and are forgotten automatically,” so an outdated fact doesn’t keep resurfacing.
- User profiles. Static and dynamic profiles for each user.
- Namespaces. Memory can be scoped to anything: a user, an agent, a project, a team or an organization.
- Document retrieval. Search over your own documents, with connectors and extractors for PDFs, web pages, images, audio and video, plus connectors for Google Drive, Gmail, Notion, OneDrive, GitHub and S3.
How it works
Supermemory’s own model, called learner-1, does the work. It “extracts and dreams on the context of every user, task, and tenant, stores it in a vector-graph database, and brings it back to the model by injecting tokens into its context in real time.” In practice, that’s a three-step loop for developers:
- Add content. Send a conversation, document or file to a namespace. Processing happens in the background, so you wait for it to finish before searching.
- Let it learn. learner-1 updates, merges, infers and forgets facts against the existing memory.
- Recall. Search memories, source passages or both, and put the results into the model’s context. Supermemory’s pricing page cites a median recall time of 187 milliseconds.
Existing customers should note that Supermemory’s older v3 and v4 APIs shut down on December 31, 2026, with a migration guide to v5.
The MCP server and plugins
Supermemory also offers a memory MCP server at mcp.supermemory.ai. You add it to ChatGPT, Claude or another MCP client, sign in with OAuth, and it can store and recall memories across your conversations, organized into “spaces.” Its plugins for coding agents such as Claude Code, Cursor, OpenCode and Windsurf give those tools persistent memory across sessions.
Where it came from
Supermemory has an unusual origin. Founder Dhravya Shah, originally from Mumbai, built it as one of 40 weekly projects after moving to the US, initially as a tool to chat with Twitter bookmarks called Any Context, TechCrunch reported. It grew into an open-source “second brain” app that, according to Supermemory’s about page, reached more than 50,000 users.
Running that app showed that memory infrastructure for AI apps didn’t really exist, and companies started asking to use the engine. Shah, then working on AI infrastructure at Cloudflare, turned it into a product. “The consumer app has since been retired; the engine is the product.”
In October 2025, Supermemory raised a seed round led by Susa Ventures, Browder Capital and SF1.vc. TechCrunch put it at $2.6 million, while Supermemory’s own announcement said $3 million. Angel investors included Cloudflare CTO Dane Knecht, Google’s Jeff Dean, Logan Kilpatrick and Sentry founder David Cramer. Shah was 19 at the time.
Pricing
Supermemory’s pricing is usage-based, with a monthly credit balance on every plan and the same rates on each. It bills only unique content, so repeats and unchanged material are free.
| Plan | Price | Notes |
|---|---|---|
| Free | $0 | $5 of credits a month |
| Pro | $19 a month | 3 team seats, unlimited end users, priority support |
| Max | $100 a month | More credits, plus a Gmail connector |
| Scale | $399 a month | Advanced PDF extraction, S3 and web crawler connectors, SOC 2, HIPAA BAA, self-hosting option |
| Enterprise | Custom | Air-gapped or dedicated deployments, SSO, uptime SLA |
What customers say
Supermemory’s pricing page quotes several founders. Harshil Mathur, founder of Razorpay, said he “tried almost everything — structured memory files, qmd etc. The only thing that works reliably is supermemory.” These are testimonials Supermemory chose, and its benchmark comparisons with rivals like Mem0 and Zep are published on its own site, so they’re best read as the company’s case rather than independent tests.
How Supermemory compares
Supermemory competes with other memory layers for AI, including Mem0, Zep, Honcho, Cognee and Letta, and with building memory yourself on a vector database. Its distinguishing traits are the graph that updates and forgets facts, the learner-1 model, broad multimodal ingestion, and a large open-source following. Our guide to connecting Mem0 to an AI agent covers one of its main rivals.
Supermemory is a tool for developers to build memory into their own products. If you want an assistant that already remembers you, Carly has its own memory system and workspace built in: it saves named memory documents that it reads and searches in later conversations, searches a knowledge base of files you upload by meaning, and keeps your to-dos, data tables and CRM in one place. The same memory comes with you whether you use Carly inside ChatGPT, inside Claude or by text, and it also labels and sorts email across all your accounts, handles advanced scheduling, and makes and answers calls and texts.
Who Supermemory suits
Supermemory is a good fit if:
- You’re building an AI app or agent that needs to remember each user across sessions.
- You want memory that works across different models and frameworks.
- You need to ingest many kinds of content, from chats to PDFs and video.
- You want the option to self-host or run it locally.
Look elsewhere if:
- You want a ready-made personal assistant rather than an engine to build with.
- You only need search over the same documents for every user, where plain retrieval may be enough.
- You need a product with a longer enterprise track record.
FAQ
What is Supermemory?
Supermemory is a memory and context engine for AI agents and apps. It extracts facts from conversations, documents and files, stores them in a graph that updates and forgets over time, and returns the right context to AI models through an API, plugins or an MCP server.
Is Supermemory open source?
Supermemory’s main repository on GitHub is MIT licensed, with about 31,600 stars. The company also runs a hosted service with paid plans.
Does Supermemory work with ChatGPT and Claude?
Yes. Supermemory’s MCP server can be added to ChatGPT, Claude Desktop and Claude on the web, and it signs in with OAuth.
Is there a Supermemory app?
Supermemory started as a consumer “second brain” app, but the company says that app has been retired and the memory engine is now the product.
Who founded Supermemory?
Dhravya Shah founded Supermemory. It raised a seed round in October 2025 led by Susa Ventures, Browder Capital and SF1.vc.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
