openviking-plugin

OpenViking Plugins

English | 简体中文

Memory auto-recall and auto-capture for hook-enabled CLI agents, powered by OpenViking.

Claude Code is fully supported. More agents are under active development.

What It Does

OpenViking Plugins gives your CLI agent persistent, cross-session, semantic memory — completely transparently:

• Auto-Recall — Before every user message, relevant memories are silently injected into the agent's context
• Auto-Capture — After every response, new knowledge is automatically extracted and stored
• MCP Tools — On-demand memory operations (search, store, delete, health check) when the agent or user needs explicit control

No extra tool calls, no manual bookmarks. The agent simply remembers.

Supported Agents

Agent	Status	Integration
Claude Code	Supported	Hooks + MCP Server
More coming soon	In Development	—

Architecture

┌─────────────────────────────────┐
│          CLI Agent              │
│  (Claude Code, ...)             │
└──────┬──────────────────┬───────┘
       │                  │
  User Prompt Hook     Stop Hook
       │                  │
 ┌─────▼──────┐    ┌──────▼──────┐
 │ Auto-Recall│    │Auto-Capture │
 │            │    │             │
 │ Semantic   │    │ Extract new │
 │ search &   │    │ knowledge & │
 │ inject     │    │ store       │
 └─────┬──────┘    └──────┬──────┘
       │                  │
       │  ┌────────────┐  │
       └─►│ OpenViking │◄─┘
          │   Server   │
          └────────────┘

Quick Start

1. Install & Start OpenViking

pip install openviking

# macOS alternative
brew install pipx && pipx install openviking

Create the config file ~/.openviking/ov.conf:

{
  "server": { "host": "127.0.0.1", "port": 1933 },
  "storage": {
    "workspace": "~/.openviking/data",
    "vectordb": { "backend": "local" },
    "agfs": { "backend": "local", "port": 1833 }
  },
  "embedding": {
    "dense": {
      "provider": "volcengine",
      "api_key": "<your-api-key>",
      "model": "doubao-embedding-vision-251215",
      "api_base": "https://ark.cn-beijing.volces.com/api/v3",
      "dimension": 1024,
      "input": "multimodal"
    }
  },
  "vlm": {
    "provider": "volcengine",
    "api_key": "<your-api-key>",
    "model": "doubao-seed-2-0-pro-260215",
    "api_base": "https://ark.cn-beijing.volces.com/api/v3"
  }
}

Start the server:

openviking-server

2. Install the Plugin (Claude Code)

/plugin marketplace add Castor6/openviking-plugins
/plugin install claude-code-memory-plugin@openviking-plugin

Then start a new Claude session — the plugin bootstraps itself automatically.

Available Plugins

Plugin	Description	Docs
claude-code-memory-plugin	Long-term semantic memory for Claude Code	README

How It Works

Auto-Recall

1. User submits a prompt → the agent's pre-prompt hook fires
2. The plugin semantically searches OpenViking for relevant memories
3. Top-ranked results are injected into the agent's context as a system message
4. The agent sees <relevant-memories> transparently — no tool call needed

Auto-Capture

1. The agent finishes responding → the post-response hook fires
2. The plugin reads the conversation transcript incrementally (only new turns)
3. New knowledge is extracted via OpenViking's AI-powered entity extraction
4. Memories are stored in the vector DB for future recall

MCP Tools (Explicit)

When you need manual control, the MCP server exposes:

• memory_recall — Semantic search across all stored memories
• memory_store — Manually persist text as structured memories
• memory_forget — Delete memories by URI or search query
• memory_health — Check OpenViking server status

Comparison with Built-in Memory

Feature	Built-in (e.g. MEMORY.md)	OpenViking Plugin
Storage	Flat files	Vector DB + structured extraction
Search	Full context load	Semantic similarity
Scope	Per-project	Cross-project, cross-session
Capacity	Limited by context window	Unlimited
Extraction	Manual	AI-powered

Configuration

All configuration lives in ~/.openviking/ov.conf. Override the path with:

export OPENVIKING_CONFIG_FILE="~/custom/path/ov.conf"

See individual plugin READMEs for agent-specific settings.

Project Structure