Active Memory
A capability in which an AI model has built-in mechanisms for writing, forgetting and prioritizing long-term information at the architecture or training stage, rather than relying entirely on external retrieval.
Core Facts
Timeline (last 90 days)
Meta built the Wonder series specifically for "agents that act on behalf of users," which includes two key features: Active Memory and Active Focus.
Muse Wonder focuses on two major features: Active Memory and Active Focus.
Active Memory means the model can carry key information about the user across conversations, without the user having to re-explain context and preferences each time.
Whether the active memory architecture claimed by Meta implies native support at the model level is not sufficiently supported by currently public information.
Native active memory generally refers to a model having built-in mechanisms for writing, forgetting, and priority management of long-term information at the architecture or training stage, rather than relying entirely on external retrieval.
Compared to passive RAG, proactive memory systems entail higher system complexity, requiring trade-offs among information filtering, storage efficiency, and retrieval accuracy.
The proactive memory paradigm allows the system to autonomously determine which content has long-term value during a conversation and persist it, and to proactively inject relevant historical context when generating responses, without requiring the user to explicitly trigger retrieval.
When new information conflicts with old memories, the memory layer must be able to identify and update them rather than holding on to outdated knowledge.
All Facts (8)
When new information conflicts with old memories, the memory layer must be able to identify and update them rather than holding on to outdated knowledge.
80%Partially VerifiedMuse Wonder focuses on two major features: Active Memory and Active Focus.
65%UnverifiedMeta built the Wonder series specifically for "agents that act on behalf of users," which includes two key features: Active Memory and Active Focus.
50%UnverifiedActive Memory means the model can carry key information about the user across conversations, without the user having to re-explain context and preferences each time.
50%UnverifiedNative active memory generally refers to a model having built-in mechanisms for writing, forgetting, and priority management of long-term information at the architecture or training stage, rather than relying entirely on external retrieval.
50%UnverifiedWhether the active memory architecture claimed by Meta implies native support at the model level is not sufficiently supported by currently public information.
50%UnverifiedCompared to passive RAG, proactive memory systems entail higher system complexity, requiring trade-offs among information filtering, storage efficiency, and retrieval accuracy.
50%UnverifiedThe proactive memory paradigm allows the system to autonomously determine which content has long-term value during a conversation and persist it, and to proactively inject relevant historical context when generating responses, without requiring the user to explicitly trigger retrieval.
50%