Technical Glossary
This glossary defines core concepts, subsystem roles, and architectural terms used across the Nimi platform.
Identity, Agents, and Cognition
- Character: The authoritative character definition stored in Realm, including durable identity, unique identifiers, and canonical lore. Characters take two forms: PersonaCharacter (general companion) and WorldCharacter (world-specific character).
- Character Source: A certified character specification issued by Realm that Runtime uses to instantiate a LocalAgent.
- World Source: Realm-managed context that defines environmental rules and lore for characters in a given world.
- LocalAgent: A local execution instance created by Runtime under a user session to handle inference, actions, and working context.
- ConversationAnchor: A durable handle assigned by Runtime. Each LocalAgent maintains a single continuous conversation stream bound to this anchor.
- Memory: Agent-specific long-term memory owned by Cognition and accessed or consolidated exclusively through Runtime.
- Knowledge: A Cognition-owned domain that currently has no active V1 ingestion, query, or public client path.
Subsystems and Product Surfaces
- SDK: The type-safe client interface (
@nimiplatform/sdk) that connects applications to admitted Runtime and Realm operations. - Kit: Shared cross-platform UI components and presets (
@nimiplatform/kit), including ready-made building blocks such as chat feeds and settings panels. - Nimi Home: The user entry portal—currently hosted by Desktop—for managing local models, viewing system status, and launching apps in protected host sessions.
- Avatar: A dedicated rendering engine for Live2D and 3D VRM pipelines that consumes Runtime presentation events for lip-sync and expressive motion.
- Runtime: The local background engine that manages local models, hardware drivers, cloud connectors, per-agent memory mediation, and protected execution channels.
- Realm: The server-side service governing global identities, social graphs, world definitions and state contracts, and creator economy rules.
Architectural Principles and Operational Terms
- State Projection: A structured, read-only view emitted by an authoritative subsystem for downstream UI rendering while persistence stays upstream.
- Capability Intent: A structured request from an app declaring the capability it needs (such as text generation or transcription), which Runtime resolves into a concrete model and execution route.
- Typed Unavailable: A structured failure response explaining why a capability cannot run right now (for example, missing drivers, unconfigured models, or an unadmitted session operation) so the app can guide the user.
- Commit Envelope: The planned structured payload specification for shared world state mutations, capturing target entities, mutation scopes, effect classes, and provenance evidence.
- Single Source of Truth: The sole subsystem authorized to own, persist, and arbitrate a given domain of state.