OpenViking

OpenViking is a self-hosted context database for developers who want agents to retain knowledge across sessions and inspect what they remember. It puts documents, memories and skills in a browsable directory structure, with editable content instead of an opaque collection of embeddings.

Agents can search within a specific project or memory directory. Generated summaries help them judge relevance before reading full documents, so retrieval can use less context. Web pages, repositories and project docs become resources alongside user preferences and task experience. Session capture archives conversations and extracts memories into Markdown that you can inspect, edit and merge.

OpenViking connects to Claude Code, Codex, Cursor, OpenClaw and Hermes, among other agent tools. Native integrations provide automatic recall and session capture; MCP clients can access memory and context tools. A browser-based Studio lets you inspect the database, and you can self-host that interface too. The server supports separate user accounts and optional access controls for resources.

The open source server uses the AGPL-3.0 license and runs in your own environment, including through Docker, without an activation key. It needs an embedding model and a vision-language model, which can run locally or through cloud services. Supported providers include Ollama, OpenAI and Volcengine. Your model choice determines whether processing stays local or uses a cloud provider. Volcengine also offers managed hosting, while a commercial self-managed edition supports deployment in your own cloud account or an offline environment.