Hologres long memory service is a fully managed, Mem0-compatible service that builds persistent memory for AI agents. It automatically extracts key information from multi-turn conversations and persists it across sessions, enabling agents to maintain consistent context and deliver personalized responses.
-
Hologres long memory service is in free public preview.
-
During the public preview, no bills are generated.
Benefits
Hologres long memory service supports AI coding, enterprise collaboration, intelligent customer service, companion chatbots, personalized education, and more. Key benefits include:
-
Ready to use with an open ecosystem: Compatible with the open-source Mem0 framework and standard RESTful APIs. Supports one-click integration through mainstream AI agent gateways such as OpenClaw and Hermes.
-
Reliable memory continuity: Memories carry over across devices and sessions, and continue to improve as the agent learns. Recall accuracy and reasoning quality are validated against an industry benchmark. For more information, see LoCoMo benchmark for the long memory service.
-
Shared memory and knowledge assets: Captures the tacit knowledge of your staff, such as the decision logic that shapes how an agent reasons and hard-won collaboration practices, and turns it into a reusable corporate asset. Memory extends from the individual to the team and the wider organization, so that members work from shared context.
-
Lower token consumption: Memory refinement automatically compresses and deduplicates historical memories, which significantly shortens the agent context. This reduces token consumption and improves response times. In AI coding scenarios, measured token savings range from 78.7% to 98.7%.
-
High performance: Supports real-time writes and updates of vector and full-text data. Memories are queryable immediately after being written, enabling real-time interaction with models. Provides fast, high-accuracy hybrid retrieval over vector and full-text data.
-
Fully managed and scalable: You do not deploy or maintain the Mem0 server, and system deployment, O&M, and troubleshooting are handled for you. Built on the Hologres cloud-native distributed architecture, the service scales to tens of billions of memory records to accommodate business growth.
Hologres long memory service vs. Mem0
|
Item |
Hologres long memory service |
Mem0 |
|
Managed service |
Fully managed by Alibaba Cloud. Pay-as-you-go. |
Self-managed. |
|
Deployment regions |
Chinese mainland and outside |
Outside the Chinese mainland only |
|
Auto scaling |
Automatic scaling on cloud-native architecture |
Manual scaling required |
|
Data scale |
Tens of billions of records |
Limited by external storage |
|
Memory CRUD |
Compatible with Mem0 API |
Basic CRUD |
|
Vector search |
Proprietary engine, millisecond-level, tens-of-billions scale |
Dependent on external vector databases |
|
Full-text search |
Native full-text engine |
Not supported |
|
Hybrid search |
Vector + full-text + graph + scalar |
Limited |
|
Search latency |
Millisecond-level |
Varies with external storage |
Billing
Hologres long memory service uses pay-as-you-go (postpaid) billing. Billing items include compute, storage, and model resources. For more information, see Billing.
Permissions
Before a RAM user can view and manage long memory services in the Hologres console, you must grant the user the required permissions. Hologres long memory service provides the following system policies.
|
Policy |
Description |
|
AliyunHologresOpenMemoryFullAccess |
Grants full management permissions on long memory services, including viewing services, releasing services, managing API-KEYs, managing IP whitelists, and managing memory sharing. |
|
AliyunHologresOpenMemoryReadOnlyAccess |
Grants read-only permissions on long memory services. A user with this policy cannot release a service, reset the API-KEY, or manage IP whitelists. |
For information about how to grant permissions, see Grant access to Hologres for RAM users.