Llongterm provides a memory layer for AI applications, enabling them to remember user interactions and maintain long-term context. Their API integrates with existing LLMs to structure and augment memory, creating more personalized and engaging AI experiences. This allows developers to build AI that acts more like a companion by remembering past conversations.
Funding
Funding not disclosed
Founders
Product
Problem
Large language models (LLMs) often lack the ability to retain context across multiple interactions, leading to generic and repetitive conversations. This absence of persistent memory hinders the creation of personalized and engaging AI experiences.
Solution
Llongterm provides a memory layer for AI applications, enabling them to remember user interactions and maintain long-term context. The platform acts as middleware between the user and the LLM, storing user messages and finding related past interactions. It then generates a "system message" containing all the relevant memories, which is passed to the LLM along with the original user message. This augmented messaging approach allows the LLM to generate responses with greater context, improving the relevance and depth of each interaction.
Target Audience
The primary users are AI developers looking to enhance the conversational flow of their applications with persistent memory and context-aware interactions.
Features
- API for creating persistent "minds" that remember user interactions
- Middleware memory that sits between the user and the LLM
- Memory enrichment that finds related past interactions
- Generation of "system messages" containing relevant memories
- Integration with existing LLMs using the system message feature
- Knowledge map to remember details over extended periods
- Self-structuring to build a dynamic taxonomy of information
- Virtual timeline to place events in time and space
- Automatic pruning and optimization for scalability
- Option to share minds between applications and services
- Human-readable design for transparency