A next-generation RAG system that uses a filesystem paradigm to manage AI agent memories, resources, and skills.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install openviking
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install openviking using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is OpenViking?
OpenViking is an open-source Context Database developed by ByteDance, specifically designed to empower AI Agents with robust memory and retrieval capabilities. Unlike traditional flat vector databases, it implements a filesystem paradigm that organizes context using URIs and tiered storage levels. This allows agents to load information on demand, moving from abstract overviews to full content based on the complexity of the query.
Integrating this tool via Openclaw Skills provides a full RAG pipeline through the Model Context Protocol (MCP). It excels at directory recursive retrieval, which significantly improves search accuracy compared to standard flat storage methods, making it an essential component for developers building sophisticated agentic workflows.
OpenViking Use Cases
- Implementing advanced AI agent memory and document Q&A systems.
- Managing large-scale knowledge bases with semantic search capabilities.
- Recursive retrieval of complex directory structures for higher accuracy.
- Automating the ingestion of PDFs, URLs, and local files into a vector database.
- Enhancing developer workflows with efficient file and documentation search within Openclaw Skills ecosystem.
How OpenViking Works
- The user initializes the environment and configures API keys for dense embeddings and VLM generation.
- The OpenViking MCP server starts, listening for requests from AI interfaces like Claude.
- Resources such as local documents or web URLs are added to the system via the add_resource tool.
- When a query is made, the system performs a semantic search across the tiered context database.
- The RAG pipeline synthesizes the retrieved documents and uses a Large Language Model to generate an informed answer.
OpenViking Setup
First, verify if the environment is ready or run the initialization script provided by Openclaw Skills:
bash ~/.openclaw/skills/openviking-mcp/scripts/init.sh
This script clones the repository and installs dependencies via uv. Next, update the ov.conf file with your Volcengine/Ark API keys. Start the server using:
cd ~/code/openviking/examples/mcp-query
uv run server.py
Finally, connect it to your AI client using the MCP transport command:
claude mcp add --transport http openviking http://localhost:2033/mcp
OpenViking Data Schema & Taxonomy
OpenViking organizes data using a URI-based filesystem paradigm to ensure structured retrieval.
| Component | Description |
|---|---|
| URI Scheme | Uses viking://resources/ paths for identifying unique assets. |
| Tiered Context | Organizes data into L0 (Abstract), L1 (Overview), and L2 (Full Content). |
| Configuration | The ov.conf file manages embedding.dense.api_key and vlm.api_key. |
| Storage | The data/ directory maintains the local vector database storage. |
OpenViking Advanced Features
- Recursive directory retrieval for significantly improved search context compared to flat search.
- Tiered context loading (L0/L1/L2) to optimize token usage and response speed.
- Support for both dense embeddings and Vision Language Models (VLM) for multimodal capabilities.
- Seamless integration with any MCP-compatible AI agent or IDE through the Openclaw Skills framework.
- URI-based resource management allowing for precise context injection and reference.
SKILL.md
Loading
Related Openclaw Skills
OpenViking MCP Server
A high-performance Model Context Protocol server providing Retrieval-Augmented Generation (RAG) and semantic search tools for AI agents.
Email Processor
A robust Gmail automation tool that triages your inbox by archiving low-priority marketing emails while surfacing critical communications.
YouTube Watcher
Fetch and process YouTube video transcripts for instant summarization and content analysis.
YouTube Watcher
An essential skill for fetching and reading YouTube video transcripts to enable AI-driven summarization and information retrieval.
Sophie Optimizer
An automated context management utility that monitors token usage, snapshots memory, and resets sessions to maintain peak AI performance.
Bio-Reabilita-Z
An AI agent specialized in physical rehabilitation protocols and the management of philanthropic health projects.
METADATA
Featured*
