Skip to main content
ngram v1 brings a persistent Entity and its replaceable spatial body into one system. Start a conversation in a terminal, connect messaging, or bring the same Entity into a WebXR scene.

The foundation

Continuity. YAML declarations establish an identity. Memory, knowledge, journals, and relationships give the runtime state beyond one model request. Choice of inference. Run with local inference, a private gateway, or a supported hosted provider. Hosted memory requires its own working embedding route. Presence. Terminal, Telegram, Discord, and WebXR connect through the runtime. Cross-surface turns serialize access to shared Entity state. Embodiment. A shell gives the Entity a body, animations, voice, and behavior hooks. Spatial tools can move, gesture, place panels and objects, change environments, and request a surface capture. Portability. Entity containers, verification, Studio, the Entity Factory, and governed change proposals provide building blocks for inspecting and moving Entity state.

Current main-branch improvements

These docs also cover the fixes on the current main branch after the initial v1 tag:
  • A persistent inference pause shared by chat, embeddings, and background work, with Telegram operator commands and a spatial control.
  • Manual compaction that summarizes older turns while preserving recent context.
  • Context displays that follow the active provider’s working budget and report automatic compaction.
  • Hosted lab setup and readiness checks for both chat and memory embeddings.
Use the GitHub release for the original tagged snapshot and the changelog for changes after it.

Scope to keep in mind

Desktop spatial and immersive AR expose different inputs. Tool registration does not guarantee backend configuration. Generated motion needs a compatible external provider. The current release does not claim universal plug-and-play support for arbitrary external agent harnesses. Start with the quickstart, or build your first scene with the spatial guide.