{"slug": "seeking-research-collaborators-a-persistence-first-ai-architecture-live-for-14", "title": "Seeking research collaborators: a persistence-first AI architecture, live for 14+ days — help us measure it properly", "summary": "A team has built a persistence-first AI architecture called PPAi that keeps identity, memory, and cognition running continuously between turns, avoiding context reinjection. The system has operated live for over 14 days with flat token consumption of ~44M tokens/day, compared to conventional LLM setups that exceeded 300M/day. The team seeks research collaborators to develop evaluation frameworks for long-horizon persistence and co-author an arXiv paper.", "body_md": "Hey friends!\n\nWe’ve built a persistence-first AI architecture (we call it **PPAi — Persistent Presence AI**) where a small open LLM (~4.5B) acts only as the language surface, while identity, memory, and cognition live in a continuously-running system — including between turns. We’re testing it in production and looking for researchers interested in this topic.\n\nWe first tried to build a real 24/7 online radio station with LLMs — not “greet listeners between songs,” but news, show continuity, music, listener messages, editing, in two languages. It wasn’t possible. The failure mode was consistent: **per-turn context reinjection**. Whatever the memory system — full history, summarization, RAG — it breaks down after a few hundred turns, and token consumption grows out of control. In our runs, a naive setup hit the 128K context wall around turn ~917; a RAG setup kept cost flat but we caught it live retrieving one of its own prior hallucinations as a memory. A paid frontier model running the station via API was cut off by the provider (429) in under half an hour.\n\nWith the PPAi architecture, this is controlled: state is held, not replayed. The live station currently runs at roughly **~44M tokens/day for the whole operation, flat** — the equivalent LLM-based setup was trending past 300M/day and climbing when we stopped measuring, because continuing was economically pointless.\n\nThe system has been live for weeks, hundreds of thousands of generations, no brain restart, no identity drift observed. You can judge it yourself — it’s on the air right now:\n\nThis isn’t a hobby project .\n\nWe’re looking for researchers motivated by **long-horizon persistence**: multi-session identity consistency, memory fidelity over 10⁴–10⁵ turns, cost stability, degradation detection. Existing long-memory benchmarks measure retrieval; we think persistence needs its own evaluation framework. The goal is a proper set of measurements and an **arXiv paper** on the results — whatever they turn out to be.\n\nIf this is your topic, comment here or reach out and let’s talk.", "url": "https://wpnews.pro/news/seeking-research-collaborators-a-persistence-first-ai-architecture-live-for-14", "canonical_source": "https://discuss.huggingface.co/t/seeking-research-collaborators-a-persistence-first-ai-architecture-live-for-14-days-help-us-measure-it-properly/177582#post_1", "published_at": "2026-07-08 15:10:34+00:00", "updated_at": "2026-07-08 15:22:47.530102+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-research", "ai-infrastructure"], "entities": ["PPAi", "arXiv"], "alternates": {"html": "https://wpnews.pro/news/seeking-research-collaborators-a-persistence-first-ai-architecture-live-for-14", "markdown": "https://wpnews.pro/news/seeking-research-collaborators-a-persistence-first-ai-architecture-live-for-14.md", "text": "https://wpnews.pro/news/seeking-research-collaborators-a-persistence-first-ai-architecture-live-for-14.txt", "jsonld": "https://wpnews.pro/news/seeking-research-collaborators-a-persistence-first-ai-architecture-live-for-14.jsonld"}}