OpenViking: A Self-Evolving Context Database for AI Agents

Introdução

OpenViking is an open-source context database specifically designed for AI agents. It unifies agent memory, knowledge RAG, and skills into a single virtual filesystem, accessible via the viking:// protocol. This innovative approach allows AI agents to browse and manage their context deterministically, using familiar commands like ls, tree, and find, rather than querying a black-box vector store. Content is processed into three tiers, L0 abstract, L1 overview, and L2 details, and loaded on demand, significantly reducing token spend and improving efficiency. Every retrieval leaves a trajectory, making debugging and observation straightforward.

Porquê usar OpenViking

OpenViking offers several compelling advantages for developing and deploying AI agents:

One filesystem for all context: Memories, resources, and skills are organized under viking:// URIs, enabling agents to locate and manipulate context deterministically, similar to a developer working with files. This provides a structured and predictable environment for agent operations.
Tiered loading cuts token spend: Each entry is processed into L0 (abstract), L1 (overview), and L2 (details) layers upon writing. Content is then loaded only as deep as the current task requires, optimizing token usage and reducing operational costs.
Directory recursive retrieval: Vector search first identifies the highest-scoring directory, then drills down layer by layer. This ensures that retrieval results arrive with their surrounding context intact, providing richer and more relevant information to the agent.
Observable retrieval: Every query preserves its directory-browsing trajectory. If a result seems incorrect, developers can trace the exact path that produced it, greatly aiding debugging and understanding of agent behavior.
Sessions become memory: After a session commits, OpenViking asynchronously extracts user preferences and agent experience, integrating them into long-term memory. This allows agents to learn and evolve over time, improving their performance and personalization.

OpenViking has demonstrated significant improvements in user memory accuracy and agent task success, as detailed in its benchmark reports.

Instalação

To get started with OpenViking, ensure you have Python 3.10 or higher installed. For a quick hands-on experience without installation, you can try the OpenViking Studio live demo.

pip install openviking --upgrade
openviking-server init      # interactive wizard: providers, models, ov.conf
openviking-server doctor    # validate setup
openviking-server           # start (background: nohup openviking-server > openviking.log 2>&1 &)

The init command guides you through provider setup, supporting Volcengine, OpenAI, Codex OAuth, Kimi, GLM, and local Ollama. The doctor command verifies your configuration, Python version, provider connectivity, and disk space.

Exemplos de Uso

Once the OpenViking server is running, you can interact with it using the ov client CLI:

ov status
ov add-resource https://github.com/volcengine/OpenViking # --wait
ov ls viking://resources/
ov tree viking://resources/volcengine -L 2
# wait some time for semantic processing if not --wait
ov find "what is openviking"
ov grep "openviking" --uri viking://resources/volcengine/OpenViking/docs/en

OpenViking also offers integrations with various AI agents, including Claude Code, Codex, OpenClaw, Hermes, LangChain, and more, allowing you to inject OpenViking recall into your agent's context and auto-commit session memory.