Context: Proactive AI Agents Replace Chatbots in Magarshak Architecture
A new AI architecture called Context, part of the Magarshak Architecture, replaces reactive chatbots with proactive goal-directed agents. The system advances shared tasks without waiting for user prompts, using three mechanisms: write-time context assembly with near-100% KV-cache reuse, composable sandboxed wisdom programs executed without further LM calls, and proactive goal stream state machines that drive conversations toward terminal states. The approach is detailed in arXiv paper 2605.23928.
Key facts
- Context is the intelligence layer of the Magarshak Architecture.
- It replaces reactive query-response chatbots with proactive goal-directed agents.
- Write-time context assembly precomputes enriched typed attributes via Groker agents.
- Context blocks are byte-identical across turns between semantic changes, enabling near-100% KV-cache reuse.
- Composable sandboxed wisdom programs form a governed library of LM-generated imperative programs.
- Programs are declaratively wired to goal types via typed stream relations and composed via phase ordering.
- Proactive goal stream state machines drive conversations toward terminal states.
- The architecture is described in arXiv paper 2605.23928.
Entities
Institutions
- arXiv
- Magarshak Architecture