Agent-native memory infrastructure that converts conversations and agent execution into structured, persistent state for production systems. It operates as a framework-agnostic library requiring a supported LLM API, with both managed cloud and self-hosted deployment options.
Project overview
On the LoCoMo benchmark, this project achieved 81.95% overall accuracy using only 4.97% of the full-context footprint (1,294 tokens per query), reducing prompt sizes while outperforming Zep, LangMem, and Mem0.
Project type
AI Agent · RAG
Use cases
Knowledge Q&A · Automation
Deployment
Refer to project documentation
License
License pending
Best for
Developers and AI engineers building LLM, datastore, and framework-agnostic agent automation systems who need to externalize and persist conversational context.
Key capabilities
Captures structured memory from conversations and agent execution in the background after each turn and makes it available for agents to recall.
Enhances memories at entity, process, and session levels with attributes, events, facts, people, preferences, relationships, rules, and skills.
Groups LLM interactions together into sessions, with support for resetting or overriding the session ID.
Allows users to use their own database for storing memories instead of the managed cloud service.
Provides a unified tool for managing account, keys, and quotas across all SDKs.
Limitations and risks
The system cannot create memories if no attribution is provided.
The free tier is rate-limited by an IP address quota; exceeding these limits requires signing up or providing an API key.
While the system can be self-hosted and run locally, inference requires an external supported LLM API.
Getting started
Install the Python or Node package, sign up via the application portal to set your MEMORI_API_KEY and a supported LLM API key (e.g., OPENAI_API_KEY) as environment variables, and call the LLM via the registered client.
Evidence and sources
GitHub project description: Memori is agent-native memory infrastructure. A LLM-agnostic layer that turns agent execution and conversation into structured, persistent state for production systems.
README: pip install memori
README: npm install @memorilabs/memori
README: Your agent forgets everything between sessions. Memori fixes that. It remembers your stack, your conventions, and how you like things done so you stop repeating yourself.
README: Memori was evaluated on the LoCoMo benchmark for long-conversation memory and achieved 81.95% overall accuracy while using an average of 1,294 tokens per query. That is just 4.97%…