Most agentic chatbots forget like goldfish or remember like hoarders.

There’s a better way.

Rant time: I’m 𝘴𝘰 tired of seeing apps that just shove the entire conversation into context (hello, long-context degradation ) or worse, store every single message in a database and call it “memory.”

It’s not memory, it’s really just data hoarding.

Real memory needs to be 𝗮𝗰𝘁𝗶𝘃𝗲𝗹𝘆 𝗺𝗮𝗶𝗻𝘁𝗮𝗶𝗻𝗲𝗱, not accumulated. It needs to reconcile contradictions, deduplicate information, and handle facts that change over time. And it needs to do all of this without blocking multi-agent pipelines or leaking data between users.

This is why I’m particularly excited about 𝗘𝗻𝗴𝗿𝗮𝗺 - our new memory service that actually treats memory as 𝘪𝘯𝘧𝘳𝘢𝘴𝘵𝘳𝘶𝘤𝘵𝘶𝘳𝘦.

This is what it does:

𝗧𝗼𝗽𝗶𝗰𝘀 𝗮𝘀 𝗺𝗮𝗴𝗻𝗲𝘁𝘀 𝗳𝗼𝗿 𝗺𝗲𝗺𝗼𝗿𝗶𝗲𝘀 You configure what information matters for your use case using natural language descriptions. Engram only extracts memories that match your topics, giving you control over what’s relevant.

𝗔𝗰𝘁𝗶𝘃𝗲 𝗿𝗲𝗰𝗼𝗻𝗰𝗶𝗹𝗶𝗮𝘁𝗶𝗼𝗻 When new information arrives, Engram retrieves related existing memories and uses LLM tool calls to decide what to do: deduplicate, rewrite to reflect updates, keep separate, or delete. Yes - a memory system needs to 𝘧𝘰𝘳𝘨𝘦𝘵 stuff in order to be actually useful.

𝗕𝘂𝗳𝗳𝗲𝗿𝘀 𝗳𝗼𝗿 𝗺𝘂𝗹𝘁𝗶-𝗮𝗴𝗲𝗻𝘁 𝘀𝘆𝘀𝘁𝗲𝗺𝘀 Buffers collect memories across multiple pipeline runs and context windows, then flush when trigger conditions are met. This is how a multi-agent system can continuously learn without blocking each other.

𝗦𝗰𝗼𝗽𝗲𝘀 𝗳𝗼𝗿 𝗽𝗿𝗼𝗽𝗲𝗿 𝗶𝘀𝗼𝗹𝗮𝘁𝗶𝗼𝗻 Project-wide memories for shared team learning, user-scoped memories with hard isolation via multi-tenancy, or property-scoped memories for soft isolation (like per-conversation). Scopes are enforced at both write and read time, so you can’t accidentally leak data between users.

𝗙𝗶𝗿𝗲-𝗮𝗻𝗱-𝗳𝗼𝗿𝗴𝗲𝘁 𝗔𝗣𝗜 Because pipelines run asynchronously, you just call the API and continue. No blocking on memory I/O, no manual background task management.

Memory isn’t a 𝘧𝘦𝘢𝘵𝘶𝘳𝘦 you add to your agent. It’s 𝗶𝗻𝗳𝗿𝗮𝘀𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲 𝘁𝗵𝗮𝘁 𝘁𝗵𝗲 𝗿𝗲𝘀𝘁 𝗼𝗳 𝘆𝗼𝘂𝗿 𝘀𝘆𝘀𝘁𝗲𝗺 𝗱𝗲𝗽𝗲𝗻𝗱𝘀 𝗼𝗻. It needs the same guarantees as your storage layer: predictable performance, hard isolation, durability, and proper lifecycle management.

By building memory directly into Weaviate, Engram inherits all of those properties. It has the same billion-scale hybrid search, multi-tenancy, and operational guarantees that we already provide to hundreds of vector database customers.

Learn more here: https://weaviate.io/product/engram?utm_source=channels&utm_medium=vs_social&utm_campaign=engram&utm_content=268014859…

Engram | Weaviate Agents | Weaviate

Source: https://weaviate.io/product/engram?utm_source=channels&utm_medium=vs_social&utm_campaign=engram&utm_content=268014859

Engram:Memory Built for AI Agents

A fully managed memory and context service purpose-built to help agents remember, learn, and improve over time

Build personalization and continuity

Remember preferences and decisions across sessions.

Extract facts and resolve inconsistencies

Turn interaction data into memories that automatically deduplicate, reconcile, and consolidate.

Shrink context windows

Send relevant and structured memories instead of raw conversations and events.

Share memory across agents

Orchestrate multi-agent systems with shared, persistent, and scoped memories.

What is Engram?

Structured memory for agentic applications

Memory shouldn’t be an ever-growing pile of context—it should be actively maintained. Engram treats memory as structured, evolving infrastructure.

Build agents that remember and get better over time

Engram is a managed memory service built on the Weaviate vector database—designed to help your applications remember, learn, and improve over time:

• ✓Extracting what matters
• ✓Resolving inconsistencies over time
• ✓Adapting to changing information
• ✓Keeping context relevant and efficient

Read the blog post to learn more

Designed to Grow with You

Launch quickly with templates

Get running in minutes with ready-to-use templates for common use cases

Secure agents with strong primitives

Scopes for data isolation when privacy matters and for context sharing when orchestration is necessary

Customizable to fit your domains

Extensible properties and composable pipelines let you shape memory and context for your domain and business needs

Get Started

Start building reliable agents with memory and context today

Integrate Engram with your agentic applications in a few simple steps and let us handle the rest.

Choose your template

Use one of Engram’s composable templates for your application

Add your data

Send user interactions or application context via Engram APIs, no preprocessing needed

Build Without Friction

Engram handles memory extraction and management, all in the background

Create trusted agents

Retrieve memory and context in real time so your agents run consistently and reliably