René Zander renezander030

Agentic Knowledge Base — Karpathy's LLM Wiki, but for any task system you already use

Last tested: 2026-05-02 — Claude Opus 4.7, TickTick adapter, qdrant 1.x, nomic-embed-text via Ollama

Series: Production AI Automation Notes — gist #3. #1: https://gist.github.com/renezander030/9069db775e494ffd2cdd5a09adf83add · #2: https://gist.github.com/renezander030/83ad49aeffa5f8749325a2b19617823f

When Karpathy's LLM Wiki post landed, I already had semantic search over my TickTick — qdrant for vectors, nomic-embed-text via Ollama, daily cron syncing the index. What was missing was the structure — the wiki layer, the agent-data note pattern, and retrieval that actually works for how an agent queries (not how I do).

Claude Code with local LLMs and ANTHROPIC_BASE_URL: Ollama, LM Studio, llama.cpp, vLLM

Native Anthropic endpoints, tool-call compatibility, and context-window sizing for local Claude Code.

Last tested: April 2026. See Changelog at the bottom.

If this saves you setup time, follow @renezander030 — practical local AI / coding-agent infrastructure notes, weekly.

Full repo with scripts: github.com/renezander030/local-ai-coding-stack

Production AI Automation Notes #1: Agent Approval Gates — JSON schema, human review, deterministic dispatch, audit logs

Production AI Automation Notes #1: Agent Approval Gates

Updated 2026-04-28 — JSON schemas, human review, deterministic dispatch, and audit logs for AI agents that touch real systems.

The most common shape of an "AI agent" demo is: model decides → model calls API → side effect happens. That works in a sandbox. It does not work the moment the agent is allowed to send an email to a customer, update a CRM record, or trigger an n8n workflow that hits production.

The pattern that does work is older than agents and is borrowed from financial systems: draft → validate → approve → dispatch → audit. Five steps, five contracts, no exceptions.

Context7 v2 — enterprise-GraphQL MCP server pattern

Updated 2026-04-28 — what changes when your MCP server backs a private GraphQL API instead of a public docs index.

Upstash/context7 is the reference MCP server for fetching up-to-date library docs. 53k stars, clean pattern: resolve a library ID, query docs, return MCP-formatted text. Works because the data is public, the schema is stable, and auth is a single API key.

That pattern breaks the moment you point an MCP server at an enterprise GraphQL API — per-tenant auth, schemas that drift, 50-type datamodels, write-side mutations. I learned that while building leanix-mcp-integration, which bridges LeanIX's enterprise-architecture GraphQL platform to Claude.

Five extensions to Context7's pattern that the enterprise case forces:

Karpathy-skills CLAUDE.md v2 — extending forrestchang's pattern with lessons from building fixclaw

CLAUDE.md

Updated 2026-04-22 — ten rules for Claude Code: four for edit-time, six for runtime.

	"""
	reddit-reader.py
	Personal read-only script that surfaces relevant Reddit posts
	from specific subreddits based on keyword matching.

	No automated posting, commenting, or voting.
	No data storage beyond current session.
	All replies are written manually through the Reddit website.
	"""

	# AI Pipeline Engine Configuration
	# Supports multiple LLM providers, budget caps, HITL approval gates,
	# and cron-scheduled pipelines.

	operator:
	channel: telegram # or slack, discord
	security:
	max_input_length: 500
	rate_limit: 10 # messages/min/user
	strip_markdown: true # prevent prompt injection via formatting

	import { router, publicProcedure } from "../trpc";

	// Import sub-routers (each in its own file)
	import { authRouter } from "./auth";
	import { userRouter } from "./user";
	import { billingRouter } from "./billing";
	import { projectRouter } from "./project";
	import { notificationRouter } from "./notification";
	import { webhookRouter } from "./webhook";
	import { analyticsRouter } from "./analytics";

	#!/usr/bin/env node
	'use strict';

	// --- Argument Parser (zero dependencies) ---

	function parseArgs(argv) {
	const args = argv.slice(2);
	const positional = [];
	const flags = {};
	for (let i = 0; i < args.length; i++) {

	package monitoring

	import (
	"context"
	"fmt"
	"runtime"
	"sync"
	"time"
	)