Nimi Documentation
Nimi is an AI open world platform. The product idea is simple: AI agents shouldn't have to be stateless tools stuck inside one request. They can live inside long-lived worlds, with identity, memory, relationships, capabilities, and rules that stay with them across sessions and surfaces.
These docs explain the product model, the ownership boundaries, and the contracts that hold across domains.
How The Docs Are Organized
Nimi is one open world platform. The platform contains several products that together make AI agents able to live in long-lived worlds: Platform (the world model itself), Runtime (AI execution), SDK (app access), Desktop and Web (user surfaces), Realm (world truth), Avatar (embodied presentation), Cognition (memory and knowledge), and Nimi Coding (the AI development methodology that ships with everything else).
Each product has its own section in these docs.
The apps/ directory in the repository contains extension apps (parentOS, Forge, shiji, overtone, and others) that showcase what the platform can do but are not part of the platform itself. Those apps are not documented here — these docs describe the platform.
What You Will Find Here
- A product model that explains why Nimi is built around worlds, not chats.
- Authority domains that name who owns which kind of truth.
- Reading paths that move from the platform model into Runtime, SDK, Desktop, Realm, Avatar, Cognition, and Nimi Coding.
- A glossary of cross-domain vocabulary used across all pages.
The Three Layers Readers Should Keep Separate
The platform splits into three layers that are easier to keep distinct in your head than to mix together.
+----------------------------------------------------------+
| Platform Model |
| World, Agent, and the six protocol primitives |
+----------------------------------------------------------+
|
v
+----------------------------------------------------------+
| Execution Substrate |
| Runtime : providers, workflows, streaming, |
| multimodal, delegation |
| Cognition : memory, knowledge, prompt serving, |
| completion |
+----------------------------------------------------------+
|
v
+----------------------------------------------------------+
| Public Surfaces |
| SDK app boundary Desktop native shell |
| Web constrained projection Realm public read path |
| Avatar embodiment authority |
+----------------------------------------------------------+- Platform model defines worlds, agents, the six fixed protocol primitives, and the rules that say who is allowed to redefine what.
- Execution substrate is how AI work actually happens: Runtime owns providers, workflows, streaming, multimodal output, and local routing; Cognition owns memory, knowledge, prompt serving, references, and completion.
- Public surfaces turn the platform into Desktop, Web, SDK, Realm, and Avatar experiences. Each public surface has its own authority boundary and its own page in these docs.
Recommended Reading Paths
| If you want to understand... | Start here |
|---|---|
| The product, the world model, and why it exists | Platform |
| The current setup and availability posture | Start |
| How AI execution is governed | Runtime |
| How apps integrate without crossing private boundaries | SDK |
| Why Desktop and Web are not equivalent | Desktop |
| Where world truth and history live | Realm |
| How embodied AI presentation is scoped | Avatar |
| Where memory and knowledge authority live | Cognition |
| The AI development paradigm and its host-agnostic package | Nimi Coding |
| The cross-domain vocabulary used in these pages | Glossary |