volcengine/openviking

OpenViking is an open-source context database built specifically for AI agents, software programs that use large language models to carry out multi-step tasks. The core problem it addresses is that when building AI agents, the information those agents need (their memory of past conversations, documents to reference, and skills they can invoke) tends to be scattered across different systems and hard to manage. Standard approaches that store everything in flat vector databases (a way of organizing text so a computer can find related passages) lose track of the big picture and make it difficult to see why an agent retrieved particular information. OpenViking organizes all this context using a file system metaphor: memories, resources, and skills are arranged in folders and files, similar to how you organize documents on a computer. This makes retrieval more structured and transparent. It uses a three-tier loading system (L0, L1, L2) so an agent only loads the context it actually needs at each moment, reducing costs. It also provides visual traces of which directories were searched, making it easier to debug when an agent pulls the wrong information. Over time, it automatically compresses conversation history and extracts long-term memory so the agent becomes more knowledgeable with use. You would use OpenViking when building AI agents that handle long, complex tasks and need reliable, organized access to growing amounts of information, for example, a research assistant, customer support bot, or coding agent. It is written in Python and requires Python 3.10 or higher, with optional Rust components for command-line tooling.