OpenViking is an open-source context database designed for AI agents, organizing knowledge, memory, and skills into a unified virtual filesystem accessed via the viking:// URI scheme. Unlike traditional "black box" embedding systems, OpenViking allows agents to navigate context using standard filesystem operations like ls, tree, and grep. The architecture implements a three-tier loading system to optimize retrieval: L0 (one-sentence abstracts), L1 (structural overviews), and L2 (full original data), enabling agents to assess relevance before consuming high-token content. The system utilizes directory-aware retrieval, employing TrieHI to resolve directory scopes before performing vector ranking, allowing queries to be limited to specific project or memory subtrees. In technical evaluations, OpenViking 0.3.22 demonstrated significant performance gains using Doubao 2.0 Pro and Doubao-embedding-vision-251215. In LoCoMo (long-conversation user memory) benchmarks, the system achieved 80–83% accuracy—a substantial increase from the 24–57% seen in native memory—while reducing input tokens by 34.3–91.0% and query latency by 58.45–66.10%. Furthermore, on the tau2-bench multi-turn task benchmark, OpenViking increased task success rates by 6.87 percentage points in retail scenarios and 11.87 percentage points in airline scenarios compared to LLMs without memory.