> ## Documentation Index
> Fetch the complete documentation index at: https://mintlify.com/volcengine/OpenViking/llms.txt
> Use this file to discover all available pages before exploring further.

# OpenViking Documentation

> An open-source context database designed for AI Agents with filesystem-based management, hierarchical context loading, and semantic retrieval.

<div
  style={{
background: 'linear-gradient(135deg, #5d56fc 0%, #bd6cfc 100%)',
borderRadius: '12px',
padding: '60px 40px',
textAlign: 'center',
marginBottom: '40px'
}}
>
  <h1
    style={{
fontSize: '48px',
fontWeight: '700',
color: '#ffffff',
marginBottom: '20px',
lineHeight: '1.2'
}}
  >
    OpenViking
  </h1>

  <p
    style={{
fontSize: '20px',
color: '#ffffff',
opacity: '0.95',
maxWidth: '700px',
margin: '0 auto 30px',
lineHeight: '1.6'
}}
  >
    The Context Database for AI Agents
  </p>

  <p
    style={{
fontSize: '16px',
color: '#ffffff',
opacity: '0.9',
maxWidth: '800px',
margin: '0 auto'
}}
  >
    Unify memory, resources, and skills through a filesystem paradigm. Enable hierarchical context delivery and self-evolving agents.
  </p>
</div>

## Why OpenViking?

In the AI era, data is abundant, but high-quality context is scarce. OpenViking solves the critical challenges developers face when building AI Agents:

<CardGroup cols={2}>
  <Card title="Context Fragmentation" icon="puzzle-piece">
    Memory in code, resources in vector databases, skills scattered everywhere — OpenViking unifies them all under a single filesystem paradigm.
  </Card>

  <Card title="Context Explosion" icon="explosion">
    Long-running Agent tasks generate massive context. OpenViking's L0/L1/L2 hierarchical loading prevents information loss while saving tokens.
  </Card>

  <Card title="Poor Retrieval Quality" icon="magnifying-glass">
    Traditional RAG uses flat storage. OpenViking's directory recursive retrieval understands complete context with global perspective.
  </Card>

  <Card title="Context Opacity" icon="eye-slash">
    Black-box retrieval makes debugging impossible. OpenViking provides visualized retrieval trajectories for full observability.
  </Card>
</CardGroup>

## Core Features

<CardGroup cols={2}>
  <Card title="Filesystem Paradigm" icon="folder-tree" href="/concepts/viking-uri">
    Organize all context (Memory, Resources, Skills) as a virtual filesystem with `viking://` URIs. Use familiar commands like `ls`, `find`, `grep` to navigate context.
  </Card>

  <Card title="Hierarchical Loading" icon="layer-group" href="/concepts/context-layers">
    Automatic L0/L1/L2 context processing: abstracts (\~100 tokens), overviews (\~2k tokens), and full details — loaded on demand to save costs.
  </Card>

  <Card title="Directory Recursive Retrieval" icon="arrows-split-up-and-left" href="/concepts/retrieval">
    Intent analysis → directory positioning → fine exploration → recursive descent. Find semantically best-matching fragments with full context awareness.
  </Card>

  <Card title="Session Management" icon="clock-rotate-left" href="/concepts/session">
    Built-in memory self-iteration. Extract 6-category memories (profile, preferences, entities, events, cases, patterns) from sessions automatically.
  </Card>
</CardGroup>

## Get Started in Minutes

<CardGroup cols={3}>
  <Card title="Quickstart" icon="rocket" href="/quickstart">
    Install OpenViking and run your first example in 5 minutes
  </Card>

  <Card title="Architecture" icon="sitemap" href="/concepts/architecture">
    Understand OpenViking's dual-layer storage and retrieval design
  </Card>

  <Card title="API Reference" icon="code" href="/api/overview">
    Explore the complete API for filesystem, search, and sessions
  </Card>
</CardGroup>

## Installation

<CodeGroup>
  ```bash pip theme={null}
  pip install openviking --upgrade
  ```

  ```bash uv theme={null}
  uv pip install openviking
  ```
</CodeGroup>

## Quick Example

```python theme={null}
from openviking import OpenViking

# Initialize client
client = OpenViking(path="./data")

# Add resources
client.add_resource("https://github.com/volcengine/OpenViking")

# Semantic search
results = client.find("what is openviking")

# Filesystem operations
client.ls("viking://resources/")
client.tree("viking://resources/volcengine/OpenViking", depth=2)

# Session management
session = client.session()
session.add(role="user", content="How do I configure OpenViking?")
session.add(role="assistant", content="You can configure OpenViking using ov.conf...")
session.commit()  # Extracts memories automatically
```

## Community & Support

<CardGroup cols={3}>
  <Card title="GitHub" icon="github" href="https://github.com/volcengine/OpenViking">
    Star the repo, report issues, and contribute
  </Card>

  <Card title="Discord" icon="discord" href="https://discord.com/invite/eHvx8E9XF3">
    Join our community for support and discussions
  </Card>

  <Card title="X (Twitter)" icon="x-twitter" href="https://x.com/openvikingai">
    Follow us for updates and announcements
  </Card>
</CardGroup>

## Integration Examples

OpenViking integrates seamlessly with popular AI agent frameworks:

<CardGroup cols={2}>
  <Card title="OpenClaw Plugin" icon="puzzle-piece" href="/integrations/openclaw">
    Boost OpenClaw task completion by 49% with 91% lower token costs
  </Card>

  <Card title="Claude Desktop MCP" icon="desktop" href="/integrations/claude-desktop">
    Connect OpenViking to Claude Desktop as a Model Context Protocol server
  </Card>

  <Card title="OpenCode Integration" icon="code" href="/integrations/opencode">
    Use OpenViking as context management for coding agents
  </Card>

  <Card title="MCP Server" icon="server" href="/integrations/mcp">
    Expose OpenViking functionality through Model Context Protocol
  </Card>
</CardGroup>

## What Makes OpenViking Different?

Traditional RAG systems treat context as flat text chunks. OpenViking treats context as a **hierarchical filesystem**:

| Traditional RAG       | OpenViking                        |
| --------------------- | --------------------------------- |
| Flat vector chunks    | Hierarchical directory structure  |
| Single-pass retrieval | Directory recursive retrieval     |
| Fixed context window  | L0/L1/L2 progressive loading      |
| Opaque retrieval      | Visualized retrieval trajectories |
| Static memory         | Self-evolving 6-category memory   |

## Real-World Performance

Based on LoCoMo10 benchmark (1,540 long-range dialogue cases):

| Configuration             | Task Completion | Input Tokens |
| ------------------------- | --------------- | ------------ |
| OpenClaw (baseline)       | 35.65%          | 24.6M        |
| OpenClaw + LanceDB        | 44.55%          | 51.6M        |
| **OpenClaw + OpenViking** | **52.08%**      | **4.3M**     |

**Results:** 49% improvement over baseline with 83% reduction in token costs.

## Open Source & Apache 2.0

OpenViking is fully open source under the Apache 2.0 license. We welcome contributions from the community.

<Info>
  **4,761 GitHub stars** and growing. Join the community building the future of AI Agent context management.
</Info>

## Next Steps

<Steps>
  <Step title="Install OpenViking">
    Follow the [Quickstart Guide](/quickstart) to install and configure OpenViking
  </Step>

  <Step title="Learn Core Concepts">
    Understand [Architecture](/concepts/architecture), [Context Types](/concepts/context-types), and [Viking URI](/concepts/viking-uri)
  </Step>

  <Step title="Explore Integrations">
    Connect OpenViking to [OpenClaw](/integrations/openclaw) or [Claude Desktop](/integrations/claude-desktop)
  </Step>

  <Step title="Deploy to Production">
    Follow the [Server Deployment Guide](/server-deployment) for production setup
  </Step>
</Steps>