OpenViking is compatible with an OSS-only privacy boundary, but this requires local inference as well as local storage. Its server supports local vector storage, local embeddings, Ollama-compatible models, and local deployment. The main project is AGPLv3.
OSS and privacy boundary
The OpenViking main project is licensed under AGPLv3; its CLI and examples use more permissive licenses. That is materially different from candidates such as MIT-licensed Hindsight or Apache-licensed Mem0 and should be reviewed if OpenViking is modified and exposed as a network service.
OpenViking can run as a self-hosted HTTP service with a local filesystem backend and local vector backend. Its embedding configuration supports a built-in local provider and Ollama; OpenAI-compatible endpoints can also point to services running on the same machine. The setup documentation explicitly provides a local-model path.
This satisfies the Operator's privacy goal only when every content-bearing model call is local. Resource semantic summaries and memory extraction require a configured VLM, and search may use a query planner or reranker. A deployment that points any of those components at a hosted API still exports project or session content even though OpenViking storage itself is self-hosted.
Sources: