Hacker
Aarush Srivastava
@aarush_srivastava
Iron Web — Shared Memory and Live Observability for AI Agents
Graph HacksOct 2026
Multi-agent applications need more than conversation history. Agents must share relevant information, track changing requirements, resolve conflicting facts, and explain which evidence informed their decisions. Iron Web provides a three-tier memory architecture: hot context for immediate instructions, source-derived context cards for specific tasks, and durable evidence stored using FalkorDB and Qdrant. Versioned claims preserve provenance and corrections, while dependency checks identify work that becomes outdated when requirements change. A separate live dashboard makes this process visible through an interactive memory graph, agent wiring, activity timeline, architecture view, and memory inspector. It displays actual application events and captured contexts. Voyara, a travel-planning application, demonstrates seven specialist agents sharing memory to produce itineraries and Travel Book PDFs. Changing a budget or introducing conflicting requirements demonstrates memory revalidation and stale-plan detection. The project is intended for developers building multi-agent applications that need inspectable, coordinated memory.
Paradox Glitch@vanshvarshney3107