index

LLM Wiki — Knowledge Base

Two-layer architecture: common/ = shared knowledge, users/ = personal notes How to use: Ingest sources into common/raw/ or users/<name>/raw/ Last updated: 2026-10-01

Structured Content

Concepts

• llm.md — Large Language Models
• ubuntu-linux.md — Ubuntu Linux
• programme-educatif-francais.md — French Educational Curriculum
• programme-francais-ce2.md — French Curriculum CE2
• programme-mathematiques-ce2.md — Mathematics CE2
• programme-questionner-monde-ce2.md — Questionner le monde CE2 (espace, temps, vivant, matière)
• programme-emc-ecole-elementaire.md — EMC école élémentaire (nouveau programme 2024, CE2 2026)
• programme-autres-matiere-ce2.md — Arts, EPS, LV1, numérique en CE2
• cours-francais-cp-3eme.md — French Language Courses CP-3ème
• cours-mathematiques-cp-3eme.md — Mathematics Courses CP-3ème
• cours-sciences-cp-3eme.md — Science Courses CP-3ème
• cours-histoire-geographie-college.md — History & Geography
• cours-anglais-cp-3eme.md — English Courses CP-3ème
• cadre-programmes-officiels.md — Official French Curriculum Framework
• sequence-pedagogique-template.md — Pedagogical Sequence Template
• guides-generation-cours-ia.md — AI Course Generation Guide
• python-programming.md — Python Programming
• docker-containers.md — Docker Containers
• git-version-control.md — Git Version Control
• cicd-devops.md — CI/CD and DevOps
• cloudflare-pages-deploy.md — Cloudflare Pages: Python deployer pattern, projects, pitfalls (no wrangler)
• honcho-local-deployment.md — Self-hosted Honcho on local llama.cpp (deriver root causes & config)
• local-llm-stack.md — Local llama.cpp stack (dual AMD GPUs, Qwen3.6-35B / Qwen3.8-27B, nomic embeddings) + vLLM ROCm test server vllm-06b (:8010, Qwen3-0.6B, GPU 0) + voice4b Qwen3-4B (:8020, GPU 1) for the voice assistant
• web-news-scraping-stdlib.md — Stdlib-only news scraping toolkit (Bing HTML/news, JSON-LD headlines, RSS/RDF/Atom freshness) + RSS-first research mode (2026-09-14: 19-outlet live-RSS list, rss_probe/rss_batch)
• hermes-cron-operations.md — Hermes cron ops: timeout limits (idle 600s, API 240s, stale floor 900s), failure modes (connection error, web_search cap, idle mid-write, response truncation, script rot), cron sandbox isolation (2026-09-25: no network + no hermes_tools), prompt-hardening patterns (preflight, atomic writes, context-bloat rules), maintenance-job recovery recipe, job authoring, gateway drop-ins (2026-09-25: +sandbox isolation entry)
• realtime-voice-assistant.md — Real-time voice assistant (voice.meiyoucheveux.com PWA, EN/ZH, barge-in): Silero VAD + faster-whisper medium + Qwen3-4B on :8020 + Piper, voice-rt.service :8090 via cloudflared ws tunnel
• pengcheng-fashion-exhibition.md — Pengcheng & Shenzhen Fashion Exhibition deep-dive: GL events Pengcheng corporate structure, market analysis, exhibition strategy (CONFIDENTIAL)

Entities

• Organizations, tools, people, places

Summaries

• Sources, papers, articles, books

Syntheses

• Cross-cutting analyses, comparative studies