OpenViking

Active
GitHub Python AGPL-3.0

Description

OpenViking is an open-source context database from Volcengine that unifies management of agent memory, resources, and skills through a filesystem paradigm, enabling hierarchical context delivery and self-evolution.

Key Features

  • Filesystem paradigm management — organize agent memory, resources, and skills in a filesystem structure, eliminating fragmented storage
  • L0/L1/L2 three-tier context loading — on-demand hierarchical context loading to significantly reduce token consumption and costs
  • Directory recursive retrieval — combines directory positioning with semantic search for recursive and precise context acquisition
  • Visualized retrieval trajectory — supports directory retrieval path visualization to help debug and optimize retrieval logic
  • Automatic session management — auto-compresses conversation content, resource references, and tool calls to extract long-term memory
  • Multi-VLM provider support — supports Volcengine Doubao, OpenAI, Kimi, GLM and other vision-language models

Use Cases

💡 Agent long-term memory management: provide cross-session persistent memory storage and retrieval for agents
💡 Multi-agent collaborative knowledge sharing: multiple agents share the same context database for collaboration
💡 Enterprise knowledge base construction: unify scattered documents, code, and resources into agent-usable context
💡 RAG system optimization: replace traditional vector databases with more efficient context retrieval and management
💡 Agent development debugging: locate context management issues through visualized retrieval trajectories

Strengths & Limitations

Strengths

  • Actively maintained, recent updates
  • High community interest (28.7k stars)

⚠️ Limitations

  • Restrictive license (AGPL-3.0)

Quick Start

```bash
pip install openviking --upgrade --force-reinstall
openviking-server init
openviking-server start
```

Related Projects

MemAgent

1.1k · Python
Stale B

A MemAgent framework that can extrapolate to 3.5M context tokens, along with a training framework for RL training of any agent workflow.

memoryagentrag +2
  • · Ultra-Long Context Processing: Extrapolate from 8K training context to 3.5M tokens with performance loss under 5%
  • · Reinforcement Learning Driven: Trained with RLVR (Reinforcement Learning from Verifiable Rewards), extends DAPO algorithm for end-to-end multi-turn conversation optimization
  • · Linear Time Complexity: Breaks through computational bottlenecks in long-text processing with linear resource scaling

Memori

16.1k · Python
Active A+

Agent-native memory infrastructure that turns agent execution and conversation into structured, persistent state with an LLM-agnostic memory layer, MCP integration, and Python/TypeScript dual SDK support.

memoryagentrag +3
  • · Agent-native memory infrastructure converting conversations and execution into structured persistent state
  • · LLM-agnostic design supporting OpenAI, Anthropic, and any model provider
  • · Python/TypeScript dual SDK with MCP one-command integration for Claude Code/Cursor/Codex

memU

14.3k · Python
Active A

A memory system for 24/7 proactive agents with MCP protocol integration, providing long-term memory management, skill storage, and proactive reasoning capabilities for continuously running AI agents.

memoryagentmcp +2
  • · 24/7 proactive memory agent that continuously captures and understands user intent
  • · File-system-like memory structure with folders, files, symlinks, and mount points for organized knowledge
  • · Significantly reduces LLM token cost with smaller context windows (~1/10 of comparable usage)

HyperDB

1.4k · Python
Stale C

A hyper-fast local vector database for use with LLM Agents, providing lightweight vector storage and similarity search capabilities for embedding as instant memory and knowledge retrieval components in agent applications.

vector-databasepythonagent +2
  • · Hyper-fast local vector database designed specifically for LLM Agents
  • · C++ high-performance backend with hardware-accelerated operations via MKL BLAS
  • · Simple unified interface compatible with all major LLM agent frameworks