From Flat Vector Soup
to Hierarchical Filesystem Brain.
OpenViking isn't a RAG wrapper. It's a context operating system for AI Agents. Every memory, skill, and resource gets a URI. Every retrieval is traceable. Every session makes the agent smarter. No ceiling on the intelligence accumulation curve.
💎 The Jewel Points — Real Alpha
Five architectural decisions that make OpenViking a category-defining system. Click each to drill down.
🏗️ The Agentic Hierarchy Stack
Three-Brain Architecture · Unified Under One Filesystem Protocol
viking:// │ ├── agent/skills/ ← Executable intelligence (tools, workflows) ├── agent/memories/ ← Accumulated task experience (self-evolving) ├── agent/instructions/ ← Behavioral constitution │ ├── user/memories/ ← Personalization layer (preferences, habits) │ └── resources/ ← World knowledge (docs, repos, web, data) └── [L0 → L1 → L2] ← Tiered loading at every node
World Knowledge
Docs, repos, web, structured data — all tiered L0→L2, all recursively searchable.
User Model
Preferences, habits, communication style — auto-extracted post-session, always current.
Agent Self-Model
Tool usage patterns, operational experience — the agent's own growing intelligence.
📊 Benchmark Proof
⚡ The Infinite Hook Decoded
The reason this ecosystem compounds infinitely:
"Every interaction writes back to the filesystem. Every write improves future retrieval. Every retrieval is observable and debuggable. The Agent's intelligence is a living, versioned, navigable artifact — not a black box."
🚀 Builder's Edge — Highest-Leverage Entry Points
VikingBot
Rapid agent scaffolding. Drop into any repo and get a context-aware agent in minutes.
OpenClaw / OpenCode
The canonical memory integration patterns. Study these first — they define the architectural contracts.
ov find + ov grep
Expose these directly to your LLM's tool-use layer. Maximum context precision, minimum token cost.