Recalld – A memory layer for AI agents that returns only relevant facts (recalld.ai)

Recalld has launched a new memory layer designed for AI agents that enhances their ability to retain and utilize relevant information over time. Unlike traditional systems that require complex setups, Recalld streamlines the process with just a single API endpoint for writing memory and two for reading it. It allows users to input conversations and documents once, with the system intelligently extracting and updating essential facts whenever new information is introduced. This significantly reduces the burden of managing vectors and memory retrieval pipelines. This innovation is crucial for the AI/ML community as it promises to minimize context loss and potential hallucinations in AI responses, thereby increasing accuracy and efficiency. Recalld employs a hybrid retrieval method paired with a language model (LLM) curation pass, which delivers targeted responses based on specific queries while conserving tokens—a significant cost-saving measure for developers. The system has been benchmarked thoroughly, demonstrating an impressive performance with 88.7% accuracy on memory recall compared to 88.2% using standard search methods, achieved with 6.7 times fewer tokens used. Additionally, Recalld ensures GDPR compliance and data security with region-specific storage and comprehensive data management features, making it a compelling option for developers looking to implement effective long-term memory solutions in AI applications.