Mem0 Self-Hosted Memory API packaged by Software Sushi

This is a repackaged open source software product wherein additional charges apply for Software Sushi packaging, maintenance and setup documentation. Mem0 OSS 2.2.1 is licensed under Apache-2.0. Software Sushi packages the original Python library with an authentication and service adapter on Ubuntu 24.04. Upstream project names and trademarks belong to their respective owners; this offer does not claim their endorsement.

Store text exactly, retrieve memories by ID, search using embeddings, or optionally extract facts with your configured chat provider. Each authenticated API user has separate local Qdrant storage and SQLite history. The API listens on HTTPS 127.0.0.1:8443 and requires authentication for every route. There is no browser dashboard, hosted-service subscription, enterprise control plane or graph-memory service included.

Recommended instance: t3.large with 8 GiB RAM and at least 24 GiB encrypted gp3 root storage. Supported minimum: t3.medium with 4 GiB RAM and 24 GiB EBS. The single-instance service processes one operation at a time and supports up to 100 configured users; it does not provide clustering, replication or automatic backups. Size storage and backups for your workload.

Connect with AWS Systems Manager Session Manager using an appropriate instance profile and outbound endpoint access; the session starts as ssm-user. Run sudo -iu ubuntu, then sushi-mem0-api GET /health. No public application port is required. Alternatively use SSH as ubuntu through your approved administrative network. Configure an OpenAI-compatible embedding endpoint and optional chat endpoint with the interactive on-instance helper, then run sudo -iu ubuntu bash /opt/sushi/docs/demo.sh. The walkthrough stores, searches and rereads synthetic facts, checks user isolation and saves results under ~/mem0-results.

Software costs USD 0.03 per running instance-hour on each offered instance type. EC2, EBS, data transfer and model-provider usage are charged separately. No provider keys or model weights are included. Memory text and search queries are sent to the provider you configure for embedding, and optional extraction sends text to its chat model. Review provider data handling and charges before using real information. Upstream telemetry is disabled.

Persisted data and history survive service restarts and reboots on the same EBS volume. Termination and EBS deletion can destroy them. Back up /var/lib/mem0 and /etc/mem0 together while stopped, and preserve your embedding provider identity. The English guide covers provider setup, user creation, recovery, limits and licensing. Original sources and dependency notices are retained on the instance.