Date: 2026-04-11
A user-facing coherent agent experience that may internally use multiple runtimes, specialists, and subsystems while appearing as one consistent entity to the user.
The shared continuity unit for an assistant interaction. An assistant session may span multiple surfaces and should survive re-entry, reattachment, and background work.
A user-facing interaction medium such as web, desktop, Slack, WhatsApp, Telegram, CLI, or voice.
A binding between a surface-specific thread/channel/window and an assistant session.
An internal agent or subsystem with a narrower responsibility than the primary assistant, such as memory synthesis, proactive watch, code review, or digest generation.
The subsystem responsible for routing work among specialists, collecting outputs, resolving conflicts, and preserving one coherent assistant response.
The focused inter-agent signaling layer used for efficient coordination between assistant subsystems and specialists. It is not generic chatter; it is structured signaling for convergence, attention, escalation, and synthesis.
The lower-level AgentWorkforce substrate that owns transport adapters, normalization, delivery, auth, scheduler substrate, relaycast/channels, and low-level action dispatch.
The process by which ephemeral conversational details are elevated into longer-lived durable memory.
The process of reducing prior context into a smaller representation that preserves useful meaning for future continuation.
A routing preference that keeps work, sessions, or surfaces attached to the runtime or environment most suitable for continuity, performance, or policy.
The subsystem that watches for triggers, scheduled conditions, stale states, or other evidence that the assistant should act without waiting for a new inbound message.
The signals used to justify proactive actions, escalations, summaries, or state transitions. Examples include stale threads, failed checks, watch hits, schedule triggers, or recent message patterns.
The open-source reusable SDK/runtime layer that defines assistant primitives without assuming hosted infrastructure.
A hosted or infrastructure-backed layer that depends on the OSS core and provides cloud adapters, managed services, or platform-specific deployment/runtime features.