Show HN: We Build Shareable Context Layer for Agent Teams (github.com)

OpenViking has launched an innovative open-source context database aimed at improving AI agents' ability to manage and utilize their knowledge, memory, and skills effectively. Unlike traditional memory systems, which often operate as black boxes, OpenViking organizes agent context within a virtual filesystem, allowing for granular access and management akin to familiar directory operations (e.g., `ls`, `grep`). This structure enhances agent navigation and usability, enabling them to inspect summaries before accessing full content, streamlining memory retrieval, and contextual understanding. This development is significant for the AI/ML community as it enhances the way agents retain and utilize knowledge, offering a robust framework for long-term memory and experience management. Early evaluations indicate that OpenViking markedly improves accuracy rates in long-conversation user memory (LoCoMo) and multi-turn tasks (tau2-bench), achieving 80-83% accuracy—up from 24-57%—while considerably reducing input tokens and query latency. These advancements promise to elevate the performance of LLMs (large language models) in real-world applications, paving the way for agents to become more adaptive and responsive as their interactions expand.