Engram: Memory by Weaviate

Engram is a managed memory service built on Weaviate, designed to give AI agents persistent, structured memory. Rather than treating memory as an ever-growing context window, Engram uses asynchronous pipelines to extract, deduplicate, and reconcile memories from raw data. Key features include configurable topics that act as filters for what information to retain, memory scopes (project-wide, user-scoped, property-scoped) enforced via Weaviate's multi-tenancy, and buffer steps that aggregate information across multiple pipeline runs. The service supports continual learning for agents, including multi-agent systems where relevant information is spread across separate context windows. Pipelines are built on Temporal for durable execution, and a Python client provides a simple fire-and-forget API for adding data and semantic search for retrieval.