WFM paper replaces sparse knowledge graphs with an 'LLM Wiki' for agent memory, trains 10.5x faster
maier_ak · x · 2026-09-21
Core of the paper
arXiv paper WFM: Wiki Foundation Model for Complex Agentic Reasoning (Junnan Dong et al., 12 authors, submitted Sep 16, 2026) proposes a new paradigm for agent long-term memory:
- Motivation: agents need persistent non-parametric knowledge; sparse graph representations limit machine readability and semantic density for complex agentic workflows. The industry is shifting to an "LLM Wiki" — dense document contexts plus markdown files with multi-layered topological linkages that agents can actually reason over.
- Method: (i) a Wiki Graph schema bridging fine-grained structure with dense contexts; (ii) query-conditioned attentive aggregation for wiki message passing with attention variance regularization; (iii) scalable representation and retrieval designed to avoid distributed overhead, claiming 10.5x faster training for large-scale commercial deployment.
More from coding & agent
- Typed Evals: open-source framework for calibrated Jev-powered LLM evaluation — Charming_Group_2950 · 2026-09-21
- diagram-design skill for Claude Code hits 41.5k GitHub stars with editorial-quality diagrams — bibryam · 2026-09-21
- Qwen-Image 2.1 open-weights 7B model gets Day-0 ComfyUI support, runs in 30s on a laptop — Lexius2129 · 2026-09-21
- Dev builds bare-metal autonomous agent system after OpenAI refused to refund lost quota — epicskyes · 2026-09-21
- Open-source 'jev' tool automates computer use with no LLM, sub-second latency — _AustinCalvert_ · 2026-09-21
- Ex-Stripe Exec: Ready Your Business for Agent Buyers With Checkout, Payment Element and Invoices — jeff_weinstein · 2026-09-21