{"slug": "infinitemem-deterministic-episodic-memory-locomo-70-49", "title": "InfiniteMem – deterministic episodic memory, LoCoMo 70.49", "summary": "InfiniteMem, a deterministic, self-hosted episodic memory system for AI agents, reports a LoCoMo QA score of 70.49% and claims a capacity of 1 million episodes or 145 million tokens per deployment, roughly 1,130 times a GPT-4o context window. The system, which is closed source and uses an MCP server for integration, is detailed in a repository by marcot3ssar1, with evaluation scripts and results available for reproducibility.", "body_md": "Deterministic, self-hosted episodic memory for agents (K3 engine):\n**1M episodes / 145M tokens per deployment** — ~1,450 novels, ~1,130× a\nGPT-4o context window, ~27 years of chat. Details: [CAPACITY.md](/marcot3ssar1/InfiniteMemOs/blob/main/CAPACITY.md).\n\nLoCoMo QA 70.49% under equal protocol, reproducible evaluation glue.\n**No implementation sources are included by design. Closed source — see NOTICE.**\n\n**Try it:** an MCP server with a free trial is coming — run infiniteMem\nfrom Claude Desktop and your agents without sharing data externally.\nContact via repository for trial access.\n\n`paper/`\n\n— scientific report v2 (LoCoMo + LongMemEval, stronger-reader rows, world-model gate, reranker study, limits, references).`harness/`\n\n— evaluation scripts only (dataset fetch checks, fresh embedding via NVIDIA API, numpy retrieval crosscheck, QA answer+judge via OpenRouter, rerank gate). They drive the closed engine as a black box and reveal nothing about its internals.`results/`\n\n— certified numbers:`leaderboard.json`\n\n(LoCoMo, SOTA2-ready),`leaderboard_lmem.json`\n\n(LongMemEval-S),`readers.json`\n\n(stronger-reader rows + world-model gate), per-question`qa_cache.jsonl`\n\n(3070 records, no gold texts), retrieval and QA reports, reranker negative-result report.\n\n- Fetch\n`snap-research/locomo`\n\n`data/locomo10.json`\n\n(CC BY-NC 4.0, non-commercial, cite Maharana et al., ACL 2024) and verify`SHA256 79FA87E9…`\n\n. - Set env keys (never commit):\n`NVIDIA_API_KEY`\n\n(embeddings`llama-nemotron-embed-vl-1b-v2`\n\n, 768-D, passage`Speaker: text`\n\n/ query raw) and`OPENROUTER_API_KEY`\n\n(answer+judge`openai/gpt-4o-mini-2024-07-18`\n\n, temp 0, verbatim Mem0 prompts). - Run harness in order:\n`exnovo_embed.py`\n\n→`eval_fresh_retrieval.py`\n\n→`exnovo_qa.py --mode k3`\n\n+`--mode oracle`\n\n. Expected: retrieval`78.85/74.10`\n\n, QA`k3 70.49 / oracle 84.89`\n\n(±0.5pp API variance).\n\nBlack-box engine, brute top-10 single-shot, dedup off, no rerank, 1 run. Answer/judge models and prompts pinned (see paper §3, §6). Dataset CC BY-NC 4.0. Contact via repository for engine access.", "url": "https://wpnews.pro/news/infinitemem-deterministic-episodic-memory-locomo-70-49", "canonical_source": "https://github.com/marcot3ssar1/InfiniteMemOs", "published_at": "2026-09-03 20:47:42+00:00", "updated_at": "2026-09-03 21:23:40.170077+00:00", "lang": "en", "topics": ["artificial-intelligence"], "entities": ["InfiniteMem", "marcot3ssar1", "LoCoMo", "GPT-4o", "MCP", "NVIDIA", "OpenRouter", "Mem0"], "alternates": {"html": "https://wpnews.pro/news/infinitemem-deterministic-episodic-memory-locomo-70-49", "markdown": "https://wpnews.pro/news/infinitemem-deterministic-episodic-memory-locomo-70-49.md", "text": "https://wpnews.pro/news/infinitemem-deterministic-episodic-memory-locomo-70-49.txt", "jsonld": "https://wpnews.pro/news/infinitemem-deterministic-episodic-memory-locomo-70-49.jsonld"}}