Browse / LLM wiki / LLmWiki-KennisBank
LLmWiki-KennisBank
Jvdbreemen
Personal LLM wiki layer for Claude Code.
Open Jvdbreemen/LLmWiki-KennisBank → Checked against the four rules.
Works with any agent
ships AGENTS.md, the cross-tool instruction file.
In its own words
Based on Karpathy's LLM Wiki pattern: raw → compile → query. Includes /sessielog, /wiki, /autoresearch, semantic tiling, and auto-crosslinking.
LLmWiki-KennisBank
Installing with an AI agent? (Claude Code, Claude Cowork, Codex, Copilot CLI, OpenCode) → follow the Agent install guide. One command per platform;
AGENTS.mdis the underlying deployment contract.
A sovereign memory layer for serious AI work.
Every agent session creates valuable context: decisions, fixes, preferences, architecture trade-offs, dead ends, and lessons you do not want to rediscover next week. Then the model forgets. KennisBank turns that temporary context into a durable local knowledge system for Claude Code, Codex, OpenCode, the GitHub Copilot CLI, and other developer agents.
It captures what happened, distils it into a sourced wiki, extracts time-aware memories, retrieves the right knowledge before the next answer, and measures whether that knowledge actually helped. The result is an AI workspace that gets sharper over time without handing your private work to a hosted memory vendor.
Plain markdown. Local SQLite. Local Ollama by default. Your own machine. You stay the editor-in-chief: the system proposes, quarantines, and flags, but a human merges, supersedes, and decides. Open the vault in Obsidian and it is just notes. Very well-organized notes that happen to power an AI memory.
The repository's own description, then 2% of its README. Read the rest on GitHub ↗
Where it came from
What this copy built: +C4-Documentation +adapters +atlas +backlog +commands +vault-structure — top-level directories this copy added on top of the original.
And a tool: +.superpowers — configuration the original does not ship.
Read from the repository's git tree on 2026-08-17.