01
Hybrid search
Reciprocal Rank Fusion over pgvector cosine + Postgres full-text. Semantic recall that still nails exact terms.
AI agent memory · Public Beta
Give your agents
Give your agents
perfect recall.
Store, search and forget long-term memories over a single API. Hybrid vector + keyword search, multi-tenant, one line of code.
preferencedark mode + compact layout0.94
Quickstart
Two calls to start. A full memory API when you grow.
Get a key, store a memory, recall it semantically — that's the two calls below. When you need more, remember turns prose into evolving facts, add_many bulk-imports thousands at once, and ready-made connectors sync your existing tools. No schema, no infra — embeddings, hybrid search and isolation are handled for you.
from longmem import Longmem mem = Longmem(api_key="mv_…") # store a memory mem.add("User prefers dark mode", category="preference")
# recall it — semantically hits = mem.search("what theme?") print(hits[0]["text"]) # → "User prefers dark mode" (0.94)
Why LongMem
Memory infrastructure, not a database you babysit.
02
Auto-embedding
Text in, vectors handled. OpenAI embeddings with a Redis cache so repeat content never costs twice.
03
Multi-tenant
API-key auth with hard tenant isolation and namespaced collections. One deployment, many agents.
04
MCP-native & SDKs
Drop-in memory for agents over MCP, plus a one-line Python SDK. Recall and capture without glue code.
05
Forget on demand
Delete by id or by semantic query. GDPR-friendly by design — memories leave when you say so.
06
Built in Germany
Hosted in the EU by 11data, a German B2B data consultancy. Privacy-minded from the ground up.
07
Connect everything
Managed connectors sync Slack, Notion, Google Drive & Linear into memory server-side — connect by OAuth or token in the dashboard. Self-host recipes and a dlt destination cover any other source.
Connect everything
Your tools, in one shared memory.
LongMem syncs your sources server-side, so every agent on the team draws from the same connected memory. Connect by OAuth or token in the dashboard — no pipeline to babysit.
Slack
Notion
Google Drive
Linear
Confluence
Gmail
HubSpot
Postgres — recipe
Synced server-side
Slack, Notion, Google Drive, Linear, Confluence, Gmail, and HubSpot pull in on their own and stay current — incremental, scoped to a collection, nothing to run.
Connect in one click
Authorize with OAuth (Slack, Notion, Google Drive) or paste a token — right from the dashboard Connections panel. Credentials are encrypted at rest.
Anything else, your way
A self-host recipe for Postgres, plus a dlt destination and bulk import for any source you can script.
Capture
Drop anything. It becomes memory.
Voice notes, meeting recordings, PDFs, screenshots, plain files — drag them onto the drop zone (or record right in the browser). Audio gets transcribed, images get described, documents get read. Everything is searchable seconds later.
● voice notes
Hit record in the browser and just talk — your voice note is transcribed and searchable seconds later. No app, no file, no typing.
audio / video
Whisper transcript, chunked & indexed — an hour of audio stays fully searchable.
pdf / docs
Per-page text extraction. Find the clause, not just the file.
images
Vision captions make screenshots and photos semantically searchable.
Pricing
Start free. Pay when your agent gets serious.
Free
$0
forever
- 1,000 memories
- Hybrid search + SDK
- MCP access
- No card required
37% cheaper
Pro
$12
/ month
- 50,000 memories
- Unlimited collections
- Image upload (1 GB)
- Priority support
Supermemory charges $19/mo for comparable limits. LongMem Pro is $12/mo — same memory, 37% less.
How it compares
LongMem vs. the alternatives
| Feature | Supermemory | Mem.ai | LongMem |
|---|---|---|---|
| Price | $19/mo | $15/mo | $12/mo |
| Memories (Pro tier) | Unknown | Unknown | 50,000 |
| Hosted in Germany | |||
| Hybrid search (vector + FTS) | |||
| Self-hosting | |||
| MCP support | |||
| Open source |
About
Agents are brilliant in the moment and amnesiac the next.
Every LLM call starts from zero. Without memory, your agent re-asks what it already knew, forgets the user's preferences, and loses the thread between sessions. Bolting a vector store onto your app to fix that becomes its own project — chunking, embeddings, hybrid ranking, multi-tenancy, decay.
LongMem is that project, finished. One API to store, search, and forget long-term memories — hybrid vector + keyword search, graph relations, per-tenant isolation, in one line of code. pip install longmem-sdk, 1,000 memories free, no card.
It is open source and self-hostable: run it on our managed cloud or your own box, bring your own model and storage, and keep your data where it belongs. Built by 11data, a German B2B data consultancy, to European data-protection standards.
One API
store · search · forget.
Open source
self-host or managed.
Free to start
1,000 memories, no card.
Built by 11data
An established German B2B data consultancy
building privacy-first AI infrastructure.
Enterprise expertise
B2B data consultancy
Years of experience building production data systems for German enterprises.
Privacy first
German engineering
Built to German data-protection standards and hosted on Hetzner infrastructure in Germany.
Open source
Community-driven
Transparent development with open-source code and a public roadmap.
Ready when you are