Open source (default)
memledger ships with an OSS-default deployment shape that needs no cloud credentials. This is what you get with pip install memledger[oss].
What you get
| Layer | Component |
|---|---|
| Vector store | PostgreSQL ≥ 14 with pgvector ≥ 0.5 |
| Chain store | JsonArrayChainStore (provenance edges live alongside the record in Postgres) |
| Embeddings | fastembed running locally — BAAI/bge-small-en-v1.5, 384 dims, ~130 MB on first run |
| Audit log | Same Postgres database |
A single pip install and a single Postgres instance. That's the shape.
When to pick this path
- Single-machine development. Stand up Postgres in Docker, point memledger at it, ship.
- On-prem / air-gapped. No outbound calls to Bedrock, OpenAI, or anywhere else. Embeddings run in-process.
- Open-source-only deployments. Apache 2.0 license, no managed-service dependency.
What's not included
- Managed Postgres backups. You operate the database. Use
pg_dump, WAL-G, or your platform of choice. - IAM-style auth. Connection strings use username/password. For IAM token rotation against Aurora, see Aurora PostgreSQL.
- Hybrid search (BM25 + vector). OSS uses pure vector retrieval. For hybrid, see OpenSearch.
- Bedrock embeddings. Local fastembed is the OSS default. Switch via
EmbeddingConfig(provider="bedrock", ...)when you want managed embeddings.
Get started
- Try it in 60 seconds → OSS Quickstart
- Single-machine dev → Docker Compose
- Production OSS on Kubernetes → Helm recipe
- AWS-native production → kagent on EKS
Path forks
| You want… | Go to |
|---|---|
| The fastest possible local demo | OSS Quickstart |
| A reproducible local stack with one command | Docker Compose |
| Self-hosted production OSS on a Kubernetes cluster | Helm recipe |
| Production agents on AWS with IAM auth + Bedrock | kagent on EKS |