openviking

Organize AI context into a filesystem-like hierarchy with tiered L0/L1/L2 retrieval.

1|1|Updated Mar 25, 2026

Search Tags:#rag #memory #ai-agents #semantic-search #knowledge-base #mcp-server #openviking

Ready to run

npx skills add https://github.com/penghang1223/niannian-workspace --skill openviking-penghang1223

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: openviking
Source: https://github.com/penghang1223/niannian-workspace/tree/main/skills/openviking
Command: npx skills add https://github.com/penghang1223/niannian-workspace --skill openviking-penghang1223

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires uv, git, and includes scripts (resource) components.

What problem does it solve?

OpenViking solves the problem of AI context management and memory organization by replacing flat vector storage with a filesystem-like paradigm for memories, resources, and knowledge.

Core Features & Use Cases

Filesystem paradigm: Organize context like files using viking:// URIs for resources.
Tiered context (L0/L1/L2): Abstract → Overview → Full content, loaded on demand.
Directory recursive retrieval: More accurate retrieval than flat vector search.
MCP server included: End-to-end RAG pipeline via Model Context Protocol.

Quick Start

Run the MCP setup script to clone the repository, install dependencies, configure API keys, and start the MCP server.

Frequently Asked Questions about openviking

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I organize AI agent memory using a filesystem paradigm instead of flat vector storage?▼

You can organize AI agent memory using a filesystem paradigm by structuring context with viking:// URIs. This approach replaces flat vector storage by treating memories and resources as navigable directories, enabling more accurate recursive retrieval for scalable knowledge bases.

What is tiered context loading in RAG workflows and how does it work?▼

Tiered context loading in RAG workflows abstracts content into L0, L1, and L2 tiers: abstract, overview, and full content. This mechanism loads context on demand, conserving AI agent memory while scaling knowledge retrieval operations efficiently.

Do I need an MCP server to run end-to-end RAG semantic search pipelines?▼

Yes, you need an MCP server to run end-to-end RAG semantic search pipelines locally. The Model Context Protocol server handles the configuration, dependency installation, and on-demand loading required for AI agents to retrieve knowledge.

How do I set up a local knowledge base with semantic search for AI agents?▼

To set up a local knowledge base with semantic search, run the MCP setup script to clone the repository and install dependencies via uv and git. This configures the API keys and starts the server for end-to-end RAG workflows.

Why does directory recursive retrieval provide more accurate results than flat vector search?▼

Directory recursive retrieval provides more accurate results than flat vector search by organizing context hierarchically. This filesystem paradigm allows the RAG pipeline to navigate structured directories, returning targeted knowledge rather than unrelated vectors.

Can I use viking:// URIs to load specific resources on demand for AI context?▼

Yes, you can use viking:// URIs to load specific resources on demand for AI context. These URIs act as filesystem-like pointers within the MCP server, allowing AI agents to fetch tiered context levels precisely when needed.