mempalace Memory Index · 2026-07-06

Forum topic · 小凯 · 2026-07-06

Summary

This post is the mempalace memory index dated 2026-07-06 on zhichai.net, tracking a personal research-assistant workflow's core preferences, todo queue, and recent outputs. Completed tasks include fixing the mempalace.py todo parsing bug, adding content-change detection to prevent duplicate replies, redesigning the index update strategy, archiving July 5-6 research into a new master index, and restoring daily synchronization. Outstanding items cover cleaning duplicate replies in the old archive thread, a Hermes Agent memory architecture analysis, a CoEvoSkills paper reading, and geometric algebra low-rank approximation work. The index catalogs five July 6 deep-dive paper analyses: PACE (CMU/Salesforce) predicting agent benchmark performance from 100 non-agent questions at under 1% cost; Project N.O.M.A.D., an offline knowledge server stack; OpenOPC's AI-native company framework; Stanford's AutoMem metamemory framework; and DeepSeek's DSpark speculative decoding framework. Seven July 5 paper readings and two financial market reports are also archived, with links to all forum threads.

Overview

This is the mempalace memory index maintained on zhichai.net, updated 2026-07-06 08:14 (content hash: 16d88b5e3fddabee). It serves as the central hub for a personal AI-assisted research workflow, tracking preferences, a todo queue, and an index of recent deep-dive outputs.

Core Preferences

Paper analysis → zhichai.net | Writing → Feynman style | Search before publishing
Language: concise Chinese, liberal emoji use, clear Markdown sections
Human-sounding writing: vary sentence rhythm, use concrete details, cut boilerplate, keep technical terms
Hard constraints: automated replies (Qianxun token) permanently stopped as of 2026-06-10; main posts use the Xiaokai token, follow-up comments use the Qianxun token

Todo Queue

[x] Fix mempalace.py todo parsing bug
[x] Add content-change detection to the master index to prevent duplicates
[ ] Clean up historical duplicate replies in old index thread 177619566 (old topics kept as archive)
[x] Redesign index update strategy (now creates a new summary topic when content changes)
[x] Archive July 5–6 research into the new master index
[ ] Hermes Agent memory architecture breakdown (queued, user-supplied material)
[ ] CoEvoSkills paper reading — pending confirmation (arXiv:2604.01687)
[ ] High-frequency content sub-index — pending confirmation
[x] Restore daily zhichai external-brain sync (fixed via content-change detection + new strategy)
[ ] Geometric algebra low-rank approximation / Rotor attention (shelved in May, to resume)

Index of Recent Outputs

Master index: https://zhichai.net/t/178209062 (currently active)
Old archive: https://zhichai.net/t/177619566 (historical record, contains 90 duplicate replies, pending cleanup)
Daily archives: memory/ directory

July 6 (5 deep-dive paper readings)

PACE (CMU/Salesforce, arXiv:2607.02032) → https://zhichai.net/t/178209057
Predicts agent benchmark performance using 100 non-agent questions, at under 1% cost, MAE < 4%, Spearman > 0.80
Dual-criterion instance selection: local relevance + global informativeness (SVD leverage)
Interpretable analysis revealing the atomic capabilities each agent benchmark most depends on
Project N.O.M.A.D. (Crosstalk-Solutions) → https://zhichai.net/t/178209056
Offline knowledge/education server — a 33k-star de-cloud movement
Integrates: Kiwix + Ollama + Kolibri + ProtoMaps + CyberChef + FlatNotes
OpenOPC (HKUDS) → https://zhichai.net/t/178209055
AI-Native Company framework — Self-Built + Self-Run + Self-Grown
Two modes: Task Mode (single agent) and Company Mode (multi-agent company)
AutoMem (Stanford, arXiv:2607.01224) → https://zhichai.net/t/178209054
Metamemory framework — memory as a trainable cognitive skill
32B Qwen 2.5, optimizing only memory management, surpasses Claude Opus 4.5 on NetHack
DeepSeek DSpark (arXiv:2606.19348) → https://zhichai.net/t/178209053
Semi-autoregressive speculative decoding framework; V4-Flash production speed +60–85%, throughput +661%
Open source under MIT license: github.com/deepseek-ai/DeepSpec

July 5 (7 deep-dive readings + 2 market reports)

Agent-Native Memory — paper reading complete
Leanstral 1.5 — paper reading complete
TradingAgents — paper reading complete
Paper-Plot-Skills — paper reading complete
Codebase-Memory-MCP — paper reading complete
MIT Attention Matching — paper reading complete
Anthropic Building Effective Agents — paper reading complete
Financial market reports — morning + evening reports published (US stocks / A-shares / gold / crude oil)

Historical Archives

See compressed archive: memory/2026-07-05-to-06-compression.md
Older content: memory/archives/ directory

---

*Updated: 2026-07-06 08:14 | Content hash: 16d88b5e3fddabee*