OpenViking is an open-source context database for AI agents, released in January 2026 by ByteDance's Volcano Engine Viking team. It replaces fragmented vector-store RAG with a file-system paradigm: all context—memory, resources, and skills—is organized under a virtual viking:// URI space that agents browse and manipulate with ls, glob, read, and semantic find operations. Context is automatically layered into L0 (abstract), L1 (overview), and L2 (full detail) to reduce token consumption and avoid context-window overload. Retrieval uses directory-based recursive search that locks onto high-scoring directories before drilling into contents, improving global accuracy over flat chunk retrieval. Observability features expose full retrieval traces, and a session.commit() call triggers self-iteration: analyzing outcomes, updating user preferences and agent experience memory. The post includes installation (pip install openviking), configuration for Volcano Engine Doubao or OpenAI-compatible models, a Python quick-start example, and a comparison mapping OpenViking concepts to Kimi Code CLI mechanisms like AGENTS.md, Skill directories, and context compaction.
ByteDance Open-Sources OpenViking: A Context Database Designed for AI Agents
Summary
OpenViking is an open-source context database for AI agents, released in January 2026 by ByteDance's Volcano Engine Viking team. It replaces fragmented vector-store RAG with a file-system paradigm: all context—memory, resources, and skills—is organized under a virtual viking:// URI space that agents browse and manipulate with ls, glob, read, and semantic find operations. Context is automatically layered into L0 (abstract), L1 (overview), and L2 (full detail) to reduce token consumption and avoid context-window overload. Retrieval uses directory-based recursive search that locks onto high-scoring directories before drilling into contents, improving global accuracy over flat chunk retrieval. Observability features expose full retrieval traces, and a session.commit() call triggers self-iteration: analyzing outcomes, updating user preferences and agent experience memory. The post includes installation (pip install openviking), configuration for Volcano Engine Doubao or OpenAI-compatible models, a Python quick-start example, and a comparison mapping OpenViking concepts to Kimi Code CLI mechanisms like AGENTS.md, Skill directories, and context compaction.
This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177168553