Self-evolving context database for AI agents
OpenViking is an open-source context database for AI agents. It unifies agent memory, knowledge resources, and skills into a single viking:// virtual file system, letting agents locate context with familiar file commands instead of querying an opaque vector store. Content is indexed into three tiers — L0 abstract, L1 overview, L2 details — at write time and loaded on demand, which cuts token usage and latency. Retrieval is directory-recursive: vector search first lands on the best-matching directory, then drills down layer by layer so results keep their surrounding context. Every query leaves an observable browsing trajectory that can be traced when answers go wrong. The project ships a CLI, server, benchmark harness, and integrations for Claude Code, Codex, Cursor, MCP clients, and more. It is well suited for engineers building agent memory, agentic RAG, or context engineering pipelines.