OpenViking

OpenViking is an open-source context database for AI agents with 40,000+ stars on GitHub, from ByteDance’s Volcengine team — one filesystem for everything an agent knows: knowledge, memory and skills.

Overview

Most agent memory is a black box: text goes in, embeddings come out, and nobody can see what was actually stored or fix it when it goes wrong. OpenViking takes the opposite approach, organizing context as a virtual filesystem under viking:// URIs that both agents and humans navigate with familiar operations — ls, tree, read, write, grep. Resources hold documents and code, memories hold user preferences and experience, and skills define how to perform tasks. Every directory carries generated summaries at two levels (L0 abstracts and L1 overviews), so agents scan summaries first and only open full content when it is relevant, and semantic search can be scoped to a project or memory subtree instead of a flat vector pool. Sessions are committed back as readable files, letting the agent’s knowledge evolve over time — hence the project’s “self-evolving context database” tagline. A browser-based Studio lets you inspect the whole store, and the project is under very active development as of October 2026.

Key Features

• One filesystem for context – knowledge, memory and skills unified under viking:// URIs.
• Transparent memory – open any directory to inspect and edit exactly what your agent knows.
• Tiered summaries – L0 abstracts and L1 overviews let agents judge relevance before reading full content.
• Scoped semantic search – search a directory subtree, not the whole index; retrieval planned from session context.
• Sessions become files – conversations commit back as readable, editable context that compounds over time.
• OpenViking Studio – visual browser interface for exploring and managing the context store.
• Agentic RAG built in – retrieval designed for agents, not bolted onto a plain vector database.

💳 Pricing

OpenViking is free and open source under the AGPL-3.0 license for self-hosting, and a hosted Studio demo is available to try in the browser. You only cover your own infrastructure and model costs.

Who Is It For

Developers building agents with long-term memory who are tired of opaque vector stores, and who want to see, debug and curate what their agent remembers.