Fetching from the wire…
Security2026-07-29 · source-backed
APPA replaces permanent context contamination with engine-managed branching: when the agent is about to acquire unvetted data, the system spawns an isolated trajectory to inspect it and returns only a bounded derivative through a trusted sanitizer to an unchanged parent context. Formally a two-monoid algebra over security labels and shared event logs. The suppression numbers are the headline but the better finding for builders is that on three of four models, branching recovered utility that ordinary taint tracking destroys. Taint tracking makes your agent dumber; branching mostly doesn't.
Each link below shares sources, entities, or timing with this story.
The diagnosis in this paper is better than the fix, and the fix is very good. Recurrent memory agents fail at long context, but not for the reason most people assume. The bottleneck isn't capture. It's retention. Retention falls below 30% at 896K tokens because every consolida...
Context inconsistency, not pattern choice, is the main reason multi-agent pipelines fail. One agent's malformed output silently corrupts everything downstream. Validate each inter-agent handoff with Pydantic or Zod before passing it on. This single discipline kills most cascad...
arXiv 2608.01326 models compaction as two games: a Context Selection Game (retain a subset) and a Context Generation Game (summarize into a bounded message). It proves the generation game is equivalent to one-way communication complexity, so the minimum compaction budget for a...
arXiv 2607.27090 targets the hidden tax in Mem0, MemGPT, and Zep: retrieved memory facts get re-prefilled on every single request. It precomputes each fact's KV representation, stores it on-GPU with a semantic embedding, and injects it into vLLM's paged cache via the KV-connec...
Three competing approaches to agent memory all trending simultaneously: - ReMe (1.7K stars) — File-first, stores as readable Markdown. Lightweight. GitHub - MemOS (6.2K stars) — Graph-based memory OS. Claims 43.7% higher accuracy vs OpenAI Memory. GitHub - Context Engineering...
Context.dev (YC S26) turns any URL into LLM-ready markdown or JSON-schema-shaped structured data, bundling JavaScript rendering, anti-bot handling, and stealth proxies at no extra credit cost, plus sitemap-discovery crawling, a Brand Intelligence API, screenshots, and NAICS/SI...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.