Weaviate has released Engram as a managed SaaS service on Weaviate Cloud, giving AI agents long-term, asynchronous memory persistence that automatically extracts and deduplicates user facts without inflating LLM context windows.
On 24 June 2026, Weaviate announced the General Availability (GA) of Engram, a managed memory and context service for agentic AI applications and large language model (LLM) workflows. It addresses production scaling limitations where AI agents struggle to reliably remember users across sessions, share multi-agent states, or learn over time without inflating prompt context windows, driving up token costs, and increasing latency.
Engram operates as background infrastructure, running fire-and-forget asynchronous pipelines to process raw events, conversation transcripts, or application metrics without introducing blocking latency to active user interactions. Instead of accumulating raw, noisy transcripts chronologically, Engram uses automated Extract, Transform, and Commit loops. It pulls facts out of interactions, deduplicates them, and continuously updates and reconciles inconsistencies against what is already known to maintain structured memory states.
Memories are natively committed directly back into the Weaviate vector database, leveraging its native multi-tenancy, multi-collection architecture, and unified hybrid search which combines semantic vector recall and BM25 keyword matching. Data isolation and sovereignty are fundamental to the architecture; memories are structurally scoped and permissioned by user ID (user_id), project, topic, or custom application properties from the primitive layer, preventing cross-tenant data leaks.
The service ships with out-of-the-box templates designed for immediate deployment around specific enterprise patterns, including personalisation, multi-agent shared states, and continual learning. The client interaction framework utilises the official Python SDK repository (weaviate/engram-python-sdk), which has been published openly on GitHub under the open-source BSD-3-Clause licence.
Engram is fully deployed as a managed SaaS solution running directly inside the Weaviate Cloud platform. It includes a Free Forever Tier providing 1,000 asynchronous pipeline runs per month, whilst paid enterprise and professional plans start at £36 ($45) per month for scalable production workloads.
