New Dynamic dreaming is now default

The context cloud for agents.

Supermemory gives your agents state-of-the-art memory, RAG, user profiles, connectors, and extractors — all built in. Extremely low latency. Works with any model.

Start Building Talk to founder
Get a personal supermemory

Used by the best teams

〉PRODUCT CATALOG [1/9]

All the legos to build the perfect context for your agent.

Focused primitives for ingesting, understanding, routing, and retrieving context.

01 · MEMORY

Memory & Continual Learning

State-of-the-art memory that is persistent, structured, and built as a knowledge graph using our custom user understanding model — powered by dynamic dreaming and a custom graph engine.

〉WHAT WE DO [2/9]

Bring your data We build understanding. Your agent just knows.

Context infrastructure for AI agents. One API, every capability.

Legacy · A vector database

Stores chunks. Returns chunks. Each session starts from zero.

  • Embeddings, namespaces, spaces
  • Append-only; no contradiction logic
  • Retrieval, not memory

Supermemory

Facts that evolve. Knowledge that merges, contradicts, and gets forgotten across every session.

  • User profiles and ontology-aware graph
  • Update, merge, contradict, infer
  • One API: ingest, retrieve, remember
For developers & teams

The Supermemory API

State-of-the-art retrieval and memory. RAG, memory, and extraction. One API, every capability.

<300ms Recall latency
100B+ Tokens / month
#1 On every bench

Self-hostable · SOC 2 Type II · TypeScript & Python SDKs

Start building in 5 minutes
For everyone

Personal Supermemory

One memory across everything you use. What you teach one AI, every AI remembers.

Supermemory App
Control panel
Claude · Cursor · Codex · OpenCode
AI plugins
Chrome Extension
One-click save
10,000+ power users
Get Personal Supermemory
〉HOW IT WORKS [3/9]

How it works

How it works.

Four primitives, one graph. Install, ingest, understand, retrieve. Then a fifth step that makes the previous four worth the effort.

quickstart.ts
TS TypeScript
PY Python
$_ cURL 
1
2
3
4
5
6
7
8
// 1. Install the SDK
npm install supermemory

// 2. Initialize the client
import { Supermemory }
  from "supermemory";

const sm = new Supermemory();
01 / PLUG IN

Plug into your stack in minutes.

SDKs for every major language and model harness. Runs in every runtime — from edge functions to long-lived servers.

02 / INGEST

Bring in any type of data, from anywhere.

Supermemory automatically extracts the content for you — in a way that's optimized for retrieval and memory generation.

03 / UNDERSTAND

Entity model learns from raw context.

Our model resolves people, places, and concepts as they evolve over time.

04 / RETRIEVE

Memories form inside our custom graph.

Memory, RAG, and profiles live in the same queryable graph — not isolated vector blobs. One unified structure your agent traverses, not three systems you stitch together.

05 / THEN — REAL-TIME TRAVERSAL

Sub-300ms graph traversal, every request.

Steps 01–04 are the build. This is the payoff — your agent reaches into the graph at request time, not at training time.

〉BENCHMARKS [4/9]

We don't think benchmarks tell the full story.

But we lead every major one anyway. SOTA on LongMemEval, LoCoMo, and ConvoMem.

We also built MemoryBench, an open eval platform for memory systems.

Recall quality,
like the human brain.

Stylized cyan-dithered illustration of a human brain

Recall time

Memories returned in milliseconds, 10× faster than Zep, 25× faster than Mem0.

3D bar chart comparing recall time: Supermemory sub-300ms vs Zep 4s vs Mem0 7-8s
Feature
Best Supermemory
Mem0
Zep
Memory Graph
Partial
User Profiles
Document Retrieval
Connectors
Document Extractors
Sub-300ms Latency
Self-hostable
Consumer Plugins
Open Eval Platform
Read the full research paper
〉USE CASES [5/9]

Use cases

Best for latency, quality and cost. Or configurable for each use case.

All use cases ↗
01 / AI Assistants

AI assistants that remember your users.

Persistent context and reasoning memory. Chats, docs, and user data unified so assistants adapt over time.

In production at
ppl.xyz endearing.aichatnow.app
Explore →
02 / Knowledge Bases

Self-improving, for your users.

Ingest, sync, and retrieve from any source -- so your agents always have the latest context, not stale snapshots.

In production at
Cluelycaret.somoonage.ai
Explore →
03 / Realtime Knowledge

Realtime knowledge for agents.

Keep your agents grounded in up-to-date facts from every source -- docs, APIs, internal tools -- synced and retrievable in milliseconds.

In production at
Composiojinso.comCluely
Explore →
04 / Internal Use

Used by engineers from the best teams.

Internal docs, wikis, and institutional knowledge -- searchable and always current for every team member.

In production at
GoogleNissan
Explore →
〉ENTERPRISE [7/9]

Enterprise

Supermemory runs everywhere.

Supermemory runs on-prem, on your cloud, or fully air-gapped — with the same API, the same SLAs, and a paper trail your security team will actually read.

01 / ON PREMISES

In your data center.

Self-host on bare metal or your own Kubernetes. Zero data leaves your perimeter.

02 / YOUR CLOUD

In your VPC.

Deploy to AWS, GCP, or Azure inside your account. BYOC from day one.

03 / LOCAL

On your laptop.

Run the full stack on a workstation for offline dev, demos, or sensitive work.

SOC2
Compliant

SOC 2 Certified

Independent audit confirming that we safeguard your data with the highest security standards.

GDPR

GDPR Compliant

Compliant with EU data protection, ensuring your personal information is handled with care and transparency.

View our security page →
〉TESTIMONIALS [8/9]

Testimonials

Armin Daryabegi

“This is crazy… We just ditched RAG completely and went memory only through @supermemory. Reduced avg response time from 40s →12s Using about 40-50% fewer tokens.”

“Just memory & near realtime web-search even for volatile information which shouldn’t read a 5 days old, cached website but 5 min old”

Chatarmin / Armin Daryabegi
12s
Memory
40s
RAG
40% Fewer Tokens Compared to RAG Baseline

“Tried almost everything mentioned here - structured memory files, qmd etc. the only thing that works reliably is @supermemory. Not facing any memory issues after setting it up.”

Razorpay / Harshil Mathur, Founder of Razorpay
Harshil Mathur

Simple pricing, by usage. Pay only for what you use.

Every plan comes with monthly credits. No surprise bills, no upgrade walls — your agent uses what your agent uses.

Free

For builders tinkering, prototypes and sideprojects.

$0 /mo
  • $5/mo of usage built in
  • Hermes Plugin
  • Supermemory MCP
  • Community support
Start free

Pro

Perfect for small teams and plugin users

$19 /mo
  • ~$20/mo of usage built in
  • Unlimited storage
  • Unlimited users
  • Auto top-up available - Buy additional usage
  • Google Drive, Notion & OneDrive connectors
  • 2 teammates included
  • OpenClaw, Claude Code and other plugins
  • Email support
Get Pro

Scale

For teams running production workloads.

$399 /mo
  • ~$600/mo of usage built in
  • Unlimited storage
  • Unlimited users
  • Up to 10 teammates
  • All connectors (Gmail, GitHub, S3, Web Crawler + Pro)
  • Auto top-up + spend caps
  • Priority support
  • SOC 2 · HIPAA BAA
  • Self-hosted option
Get Scale

Enterprise

For organizations with committed spend.

Custom
  • Unlimited usage
  • Enterprise MCP
  • Committed-spend pricing
  • Dedicated account manager
  • Air-gapped self-hosting
  • Uptime SLA
  • Custom contracts & DPA
Talk to sales
See full rate card and FAQ
〉FAQ [9/9]
FREQUENTLY ASKED QUESTIONS

The fine print, in plain English.

  1. Every plan comes with a monthly dollar balance. Each API call, storing memories, searching, indexing, draws from that balance at the rates listed above.

    When the balance runs out, you can top up or auto top-up to keep going. No surprise bills at the end of the month.

  2. SM tokens are the unique tokens Supermemory actually ingests and embeds; repeats and unchanged content don't get billed again. Effectively a 100% discount on what a normal prompt cache would re-charge you for.

    Plain text is $0.005 per 1K SM tokens; rich content (PDFs, audio, video) is $0.010 per 1K because it needs heavier extraction.

    PLAIN TEXT $0.005 / 1K
    RICH CONTENT $0.010 / 1K

  3. Nothing, you're not billed again. Supermemory deduplicates at the token level, so re-uploading a doc, syncing a connector, or pushing the same conversation history won't redraw from your balance.

    Only net-new content counts as SM tokens. This is why production agents that loop over the same context end up an order of magnitude cheaper here than with a typical vector DB.

  4. Subscription credits reset monthly. Top-up credits you purchase in advance never expire, they sit in your balance until you use them.

  5. On Free, you'll be paused; pay-as-you-go isn't available on Free, so you'll need to upgrade to Pro or Scale to keep going.

    On Pro and Scale, auto top-up kicks in to keep your app running. You can set hard spend caps on Scale to prevent runaway usage.

  6. Self-hosted deployments are available on Scale and Enterprise. Enterprise additionally supports fully air-gapped deployments (LLM inference may be the only outbound dependency, unless GPUs are available).

  7. Yes. Qualifying early-stage startups and academic research teams get $1,000 in credits, dedicated support, and 6 months to build. Apply via the Startup Program link above.

Still something on your mind?

Read the docs Talk to founder

Your Agent Second Brain Workflow needs it’s Supermemory

Start Building
TOTAL MEMORIES SAVED 1,505,650,017
22.7K stars