Production memory that scales with your agents.
MemoryLayer gets you started. Enterprise takes you to production — Postgres at the core, hybrid retrieval, knowledge graph, and the governance controls your organization requires. Open source and self-hosted.
Same engine. Built for shared production.
MemoryLayer is not a teaser — it has the whole retrieval engine: hybrid search, the typed knowledge graph, reflection, temporal recall. Enterprise is what you move to when that engine has to run as shared production infrastructure: a durable backend, governance, and the data-platform surface around it.
MemoryLayer
Apache-2.0 · Embed anywhere- SQLite backend — zero config, runs anywhere
- Hybrid BM25 + vector retrieval and the full 85-type knowledge graph
- Reflection, consolidation, temporal recall, context sandbox
- Python, TypeScript & Go SDKs, MCP server, Claude Code plugin
MemoryLayer Enterprise
AGPL-3.0 · Production · Governance- PostgreSQL + pgvector — durable, replicated, production-grade
- Everything in MemoryLayer, running at production scale
- RBAC, SSO/OIDC, audit trails, multi-tenant isolation
- Hot/warm/cold memory tiering for cost-efficient scale
- Collections, datasets, document chat, trajectories, data connectors
Everything production memory demands.
A complete memory infrastructure stack — not a feature list, a coherent system built for agents that need to remember at scale.
PostgreSQL + pgvector
Durable, battle-tested vector store. Drop SQLite when you need real persistence, replication, and query power.
The full OSS retrieval engine
Hybrid BM25 + vector retrieval, the 85-type knowledge graph, reflection, consolidation, and temporal recall — all of it, running on Postgres instead of SQLite.
Collections & datasets
A general vector-search collections API plus tabular dataset upload, profiling, and slicing — beyond the memory API surface.
Document chat
Chat directly over ingested documents, with page-level grounding and image-aware retrieval.
Trajectories
Inspect exactly why recall returned what it did — trace every retrieval arm and ranking decision when tuning production behaviour.
Hot / warm / cold tiering
Cost-efficient lifecycle management for memories at scale. Frequently accessed data stays hot; older data tiers down automatically.
RBAC, audit trails, SSO/OIDC
Fine-grained role-based access, full audit log, and SSO/OIDC integration via the Aether auth proxy.
Multi-tenant isolation
Secure per-tenant data separation at the storage and query layer. One cluster, many customers — no bleed.
Admin & platform surface
Cross-workspace stats, storage usage, job inspection, user and application management — the operational view for running memory as shared infrastructure.
Multimodal & document ingestion
Text, images, PDFs, and Office documents at scale — per-page memory extraction with LLM-driven transcription, plus chat over your documents.
Data connectors
Cloud, SaaS, and team-chat sources feed the ingestion pipeline automatically — the same SDKs and MCP server as OSS on top.
Powered by MemoryLayer Storage
Heavy document and tensor workloads get content-addressed dedup + compression under the hood.
Learn about StorageYour application code doesn't change. Only the backend does.
The MemoryLayer client SDK is identical across MemoryLayer and Enterprise. Enterprise adds the Postgres backend, hybrid retrieval, governance controls, and scale — your agent's API calls stay exactly the same.
- Python SDK: pip install memorylayer-client
- TypeScript SDK: npm install @scitrera/memorylayer-sdk
- MCP server works with Claude, Cursor, and any MCP-compatible tool
from memorylayer import SyncMemoryLayerClient
with SyncMemoryLayerClient(
base_url=MEMORYLAYER_URL,
workspace_id="my-workspace",
) as ml:
# Same call on either edition.
# Enterprise runs it on Postgres +
# pgvector, at production scale,
# under your governance controls.
results = ml.recall(
"last week's project decisions",
) Self-host it your way.
Run it from source, deploy the containers, or get it pre-integrated as part of the Scitrera Platform. Your infrastructure, your data.
Run from source
Clone the repository and bring up a combined development environment with the pinned MemoryLayer core in one command.
python3 scripts/dev.py enterprise - Python 3.12+; Node.js 22+ for the admin dashboard
- OSS core pinned to a reviewed MemoryLayer commit
- Or point it at your own MemoryLayer checkout
Containers
Multi-architecture images (amd64 and arm64) built from the repository and published to GHCR.
ghcr.io/scitrera/memorylayer-enterprise - Enterprise service and worker
- Data connectors service
- PostgreSQL 17 with pgvector, Apache AGE, and pg_textsearch
In the Scitrera Platform
Enterprise is the memory, document, and retrieval service for each tenant of the open-source Scitrera Platform.
docker compose · helm - Per-tenant instance alongside auth, agents, and storage
- Docker Compose for development, Helm for Kubernetes
- Wired to MemoryLayer Storage and data connectors
Why AGPL-3.0?
The AGPL keeps the production edition open for everyone. You can run, study, modify, and self-host MemoryLayer Enterprise freely. If you offer a modified version to others over a network, you share those modifications under the same license.
The client SDKs and MCP server your applications use remain Apache-2.0, as does the MemoryLayer core.
Scitrera has closed its doors and opened the source. There is no commercial licensing or paid support; questions and contributions go through GitHub.
Ready for production-grade memory?
Clone it, run it, and put it behind your agents. Start with MemoryLayer and move to Enterprise when you need Postgres and the production platform.
Uses MemoryLayer Storage (AGPL-3.0) for document & tensor workloads · Learn more