Supermemory: A Memory and Context Engine for the AI Era

TL;DR: supermemoryai/supermemory is the TypeScript, MIT project behind supermemory.ai — a fast, scalable memory and context engine for AI apps that can run fully locally or on Cloudflare. 30,303 stars on GitHub.

What it is

supermemoryai/supermemory describes itself as “Memory and context engine + app that is extremely fast, scalable, and can be run fully locally. The Memory API for the AI era.” That sentence positions the project at the seam between two trends: AI agents that need persistent memory across conversations, and the Cloudflare platform that has become the default runtime for shipping that kind of service. The repository is written in TypeScript, has 30,303 stars, and is published under the MIT license. The homepage field is https://supermemory.ai/docs, which is the documentation site. The GitHub topic list leans hard on the Cloudflare stack: agent-memory, ai-memory, cloudflare-kv, cloudflare-pages, cloudflare-workers. The repository was created on 2024-02-27, which makes it about two and a half years old.

The shape is an API plus a deployable reference app. The API accepts context, indexes it, and serves it back to the agent on demand. The app on supermemory.ai is the human-facing surface that exercises the same API.

What it does well

The interesting question for any “AI memory” tool is what makes it different from a vector database and an LLM call. The topic list and the description point at four answers.

• Memory as a first-class API. Both agent-memory and ai-memory are explicit topics, and the GitHub description literally calls it “The Memory API for the AI era.” That is the product positioning — a service you wire into an agent, not a database you build a memory layer on top of.
• Cloudflare-native deployment. cloudflare-kv, cloudflare-pages, and cloudflare-workers are all explicit topics, which means the reference deployment ships on the same primitives. Workers for compute, KV for the indexed store, Pages for the static front-end.
• Locally runnable. The description is explicit that the engine “can be run fully locally” — useful for development, for air-gapped environments, or for users who do not want their data on Cloudflare.
• Scalable by construction. The topic list and the description both lean on scalability, which is the practical concern once an agent starts accumulating long conversation histories.
• MIT license. No friction on embedding the API into commercial products, which matters for any team shipping a paid agent.

The 30,303-star count at 140 stars today reflects a project that has settled into a real niche: the AI-engineer crowd that needs a memory layer and would rather call an API than build one.

How to try it

The fastest path is the hosted API, which is the same engine that powers supermemory.ai.

# Install the client
npm install supermemory

# Send context into the engine
const memory = new Supermemory({ apiKey: process.env.SUPERMEMORY_KEY });
await memory.add("User prefers dark mode and vim keybindings.");

# Retrieve it on the next agent turn
const ctx = await memory.search("display preferences");

For a local install, the README on the repository walks through the same engine without the Cloudflare dependency.

# Clone and run locally
git clone https://github.com/supermemoryai/supermemory
cd supermemory
npm install
npm run dev

When it fits, and when it doesn’t

Supermemory is the right tool when your agent needs to remember things across sessions, when you would rather call an API than operate a vector database, and when the Cloudflare runtime is in scope. It is a strong fit for product teams shipping consumer agents, for prototypes that need a memory layer today rather than next quarter, and for anyone who wants the same primitive locally and in production. It is less attractive for workflows where a single conversation contains all the context, for cases where you already have a vector store you are happy with, or when the latency budget is tighter than a network round-trip to Cloudflare. The MIT license means there is no commercial friction.

The 30,303-star count, the two-and-a-half-year-old repository, and the Cloudflare-heavy topic list converge on the same conclusion: a memory service that has carved out a real place in the AI-agent stack.

Source and Accuracy Notes

• Project page: https://github.com/supermemoryai/supermemory
• Source repository: supermemoryai/supermemory (verified via GitHub API html_url)
• License: MIT
• Stars at time of writing: 30,303
• Language: TypeScript
• Topics: agent-memory, ai-memory, cloudflare-kv, cloudflare-pages, cloudflare-workers
• Homepage: https://supermemory.ai/docs
• Created: 2024-02-27