Skip to content

Mem0 Self-Hosted Memory API

AI Agents Mem0 2.2.1 Hardened AMI

Mem0 OSS 2.2.1 on Ubuntu 24.04 stores application memories and searches them with embeddings from a provider you configure. The API is authenticated HTTPS on loopback. Each API user has a separate local Qdrant store and a SQLite history. The default add stores your text as written. Optional infer: true asks your chat model to extract facts first.

This image does not include Mem0's hosted dashboard, enterprise control plane, graph memory, or a managed-service subscription. No model weights or provider keys are included.

AMI release
0.1.0 (Mem0 2.2.1)
Recommended
t3.large
Storage
24 GiB+ encrypted gp3
Access
Session Manager or SSH
Released
Oct 30, 2026
Delivery
Amazon Machine Image in your own AWS account

Store exact application memories and search them with your configured embedding provider.

Authenticated loopback HTTPS API with separate local Qdrant storage and SQLite history per API user.

English setup guide and a synthetic store, search, isolation and readback walkthrough.

Quick start

Full guide →
  1. Subscribe
  2. Launch the instance
  3. Connect
  4. Configure a provider
  5. Store and search
  6. Run the walkthrough

What is on the image

This guide matches Marketplace package 0.1.0 (Mem0 2.2.1). Check the listing for the version currently on offer.

  • Mem0 OSS 2.2.1 with a Software Sushi authentication adapter.
  • HTTPS API at https://127.0.0.1:8443. Every route, including health, requires authentication. There is no browser dashboard.
  • Per-user storage under /var/lib/mem0/users.
  • Credentials in ~/.config/sushi/mem0/access.json. The client trusts the per-instance certificate and does not put secrets on the command line.
  • An isolation-check account at ~/.config/sushi/mem0/isolation.json, used by the demo to show that another user cannot read the memory.
  • A walkthrough, /opt/sushi/docs/demo.sh.

The service handles one operation at a time and supports up to 100 configured users. It is a single instance, with no cluster replication and no automatic backups. Upstream Mem0 telemetry is disabled.

t3.large (8 GiB of memory) with an encrypted gp3 root volume of at least 24 GiB. The listing also offers t3.medium (4 GiB) as a supported minimum, plus t3.xlarge and t3.2xlarge. Use Ubuntu 24.04 x86_64. Leave inbound application ports closed. Forward port 8443 only for a client that is not running on the instance.

The public guide says to choose the Region your subscription supports. It names us-east-1 as the launch example. Confirm the Regions on the listing.

Scope and limits

POST /memories accepts text of 1 to 8,000 characters and optional infer (default false). POST /search accepts query and optional limit from 1 to 100. GET /memories lists up to 100 memories. Search returns similarity results. An id lookup returns the exact stored text. The API user comes from authentication. Request-supplied tenant selectors and arbitrary filters are rejected.

Memory text and queries are sent to the configured provider for embedding. infer: true also sends text to the chat model. Once vectors exist, the adapter refuses a change of provider URL, embedding model, or dimensions.

Billing

The listing states USD 0.03 per instance-hour. Confirm the offer, including its free trial, before you subscribe. EC2, EBS, data transfer, and model-provider usage are billed separately. The demo requires a configured embedding provider and can incur provider charges.

Where to go next

Upstream project

Mem0 OSS is Apache-2.0. Software Sushi packages it independently. Sources are in /opt/mem0/sources, wheels in /opt/mem0/wheels, and notices in /usr/share/doc/mem0. See the Mem0 open-source documentation and the public getting-started page.

Last updated November 11, 2026 · Suggest an improvement