OpenViking

Overview

OpenViking is an open-source context database for AI agents that organizes knowledge, memory, and skills as a virtual filesystem under viking://. Agents can browse, search, and read context as files, with generated summaries and layered access. It connects with popular agent harnesses and provides benchmarks showing improved long-term memory accuracy and token efficiency.

Key Features

• One filesystem for knowledge, memory, and skills: resources, user memories, and skills each get a viking:// URI for browsing and retrieval.
• Directory-scoped semantic search that runs queries within project or memory subtrees instead of scanning a flat vector index.
• Layered context with L0 abstracts, L1 overviews, and L2 full details so agents can judge relevance before reading source content.
• Sessions become inspectable Markdown files, and ov compile can organize source material into a wiki, knowledge graph, or report.
• Integrations with agent harnesses including Claude Code, Codex, Cursor, OpenClaw, Hermes, OpenCode, and MCP clients.
• Benchmarks report 80%+ long-term memory accuracy across agent frameworks and up to 63% reduction in input tokens versus baselines.

Use Cases

• Providing cross-session memory for coding agents such as Claude Code, Codex, and Cursor through hooks, plugins, or MCP.
• Organizing and retrieving knowledge from project documentation and GitHub repositories for retrieval-augmented generation.
• Improving long-term memory accuracy and reducing token costs when connecting agent frameworks like OpenClaw, Hermes, and Claude Code.
• Building custom agent integrations with Python, Go, or TypeScript SDKs and the HTTP API.

Getting Started

• Install the package with pip install openviking --upgrade.
• Run openviking-server init to configure providers and models.
• Run openviking-server doctor to check configuration and connectivity.
• Start the server with openviking-server.
• Use the ov CLI to add resources, run searches, and inspect context with commands like ov add-resource, ov find, and ov grep.

Deployment & Requirements

• Requires Python 3.10+ and access to an embedding model and a VLM, either cloud or local.
• The open-source server runs under AGPLv3 and requires no activation key; configure authentication before exposing it beyond localhost.
• A Docker and deployment guide is available, and a desktop app beta supports macOS and Windows x64.

Before You Adopt

• License: AGPL-3.0. Review its terms before using, modifying, or distributing the project.
• The server requires an embedding model and a VLM, which can be cloud or local external dependencies.
• The desktop app is currently in beta and limited to macOS and Windows x64.
• The self-managed commercial edition is activated by a license key and adds distributed deployment and official support.