What is OpenViking?
OpenViking is an open-source contextual database for AI agents, replacing flat vector storage with a file system model. It organizes context (memories, resources, skills) under `viking://` URIs with three layers: L0 (~100 tokens), L1 (~2k tokens), L2 (full content).
Why it matters
OpenViking's novel approach to contextual storage and retrieval can significantly improve the performance and capabilities of AI agents.
Key Points
- 1OpenViking unifies fragmented context into a virtual file system
- 2It introduces a hierarchical L0/L1/L2 loading model to reduce token costs by 91%
- 3The file-based approach enables efficient retrieval, observability, and reusability
Details
Traditional retrieval-augmented generation (RAG) systems store context data in flat vector databases, leading to fragmented memory, high token costs, and poor retrieval performance. OpenViking, developed by ByteDance, takes a new approach by replacing flat vector storage with a file system model. It organizes all context under unique `viking://` URIs, with the ability to load content in a hierarchical L0/L1/L2 manner. This reduces token usage by 91% and improves task completion rate by 43% compared to traditional RAG. The file-based structure also enables better context retrieval, observability, and reusability for AI agents.
No comments yet
Be the first to comment