Open source · context infrastructure
The Context Database for AI Agents
OpenViking (ov)
OpenViking unifies agent memory, knowledge RAG, and skills in one self-evolving context layer — organized like a filesystem, deployed inside your own environment. Open-source core, with a private-delivery Enterprise edition for teams that need licensing, controls, and support.
$ pip install openvikingContext, treated as infrastructure
Agents outgrow prompt stuffing and fragmented vector stores. ov gives them one consistent way to read, write, and retrieve everything they know.
One filesystem for context
Memory, knowledge, and skills live as files under viking:// paths — a single namespace instead of scattered stores and glue code.
Tiered loading, fewer tokens
Context loads in layers — abstract, overview, details — so agents read only what the task needs and token spend stays flat as knowledge grows.
Memory that compounds
Sessions distill into durable memory automatically. On the LoCoMo benchmark, agent accuracy rises from 57% to 80% — agents get smarter with use.
Data boundary
Your context never leaves your environment
Licensing verifies an ES256-signed lease locally — no agent content is sent to us. The only thing that crosses the boundary is a minimal, preference-gated operational signal.
Editions
Start open, upgrade when you need it
Community
The full open-source core under AGPL-3.0, self-hosted in your own environment. Free, forever.
Get started on GitHubEnterprise
Private delivery with a signed license, management console, operational telemetry, and support — activated with a key.
See pricing & licenseGive your agents a memory you control
Get a trial key with your email, download ov, and activate it in your own environment in minutes.