# OpenViking

OpenViking은 AI 에이전트를 위한 오픈소스 컨텍스트 데이터베이스로, 파일시스템 패러다임(viking:// URI)을 사용하여 메모리, 리소스, 기술을 구성하고 계층적 컨텍스트 전달, 계층적 로딩, 자기 진화 에이전트를 가능하게 합니다.

Canonical URL: https://a2a-mcp.org/entry/openviking-ko

Language: ko

Category: Frameworks & SDKs

Updated: 2026-04-08

## What is OpenViking?

OpenViking is an open-source **context database** specifically designed for AI agents. Developed by Volcengine (ByteDance's cloud team), it replaces fragmented vector stores and flat context pools with a clean **filesystem paradigm**.

Everything — memories, resources, and skills — becomes a file or directory accessible via `viking://` URIs. This enables hierarchical organization, layered context loading (L0/L1 lightweight + on-demand L2), semantic retrieval, and self-evolving agent behavior.

## Core Features

- **Filesystem Paradigm**: Organize agent context like a traditional file system with clear hierarchy (resources/, memories/, skills/, etc.).
- **Layered Context Loading**: Use lightweight summaries (L0/L1) by default and fetch detailed content only when needed, dramatically reducing token consumption (up to 96% savings reported).
- **Hierarchical Context Delivery**: Agents navigate context naturally with stable, observable structure.
- **Self-Evolving**: Agents can iterate on their own memory and knowledge base over time.
- **Semantic Retrieval & RAG**: Built-in support for high-performance vector indexing and retrieval.
- **MCP Integration**: Provides Model Context Protocol tools for seamless connection with Claude Desktop, Claude CLI, and other MCP-compatible clients.
- **Multi-Language Support**: Core in Rust with Python bindings and CLI tools.

## Design Philosophy

- Context as a navigable filesystem instead of opaque vector chunks.
- Minimalist interaction paradigm for scalable, long-running agents.
- Improved observability, debuggability, and maintainability of agent memory.
- Designed to work natively with agents like OpenClaw and other agent frameworks.

## Use Cases

- Building long-term memory for autonomous AI agents.
- Reducing context window costs in production agent systems.
- Creating self-improving agents that evolve their knowledge base.
- Integrating with MCP-compatible tools (Claude Desktop, etc.).
- Managing complex agent workflows involving memory, tools, and external resources.

## Quick Start

Install the CLI:

```bash
curl -fsSL https://raw.githubusercontent.com/volcengine/OpenViking/main/crates/ov_cli/install.sh | bash
```

Or build from source with Cargo.

Basic usage involves initializing a context store, adding resources/memories, and querying via the filesystem-like interface or MCP server.

Full documentation and examples are available in the repository.

## Resources

- GitHub Repository: https://github.com/volcengine/OpenViking
- Official Website: https://openviking.ai/
- Documentation: https://mintlify.com/volcengine/OpenViking

OpenViking is Apache 2.0 licensed and actively maintained, with strong community adoption for next-generation AI agent development.

## Tags

- ai-에이전트
- 컨텍스트-관리
- 메모리The request you have submitted contains the instruction to return ONLY a valid JSON object, no markdown fences, no extra text. However, your response begins with a markdown code fence (

## Sources

- [Project website](https://github.com/volcengine/OpenViking)