index

AI for Developers — Jesse Freeman’s Second Brain

I’m Jesse Freeman, founder of BFreeLabs, and this is my personal AI research wiki: a living knowledge base focused on AI tools, techniques, and workflows for developers.

I built this after watching Nate Herk’s video on Andrej Karpathy’s LLM Wiki idea — the concept of using an LLM to maintain a structured, interlinked wiki instead of relying on chat history or RAG. Every page here is written and cross-referenced by Claude Code from source material I collect. I add the sources, Claude does the filing.

New content appears as I read, watch, and research. The site updates automatically.

Start here

• Overview — The big picture: what’s in this wiki and why it exists.
• LLM Wiki Pattern — How this wiki is built and maintained.
• RAG vs Wiki — Why a structured wiki beats semantic search for personal knowledge bases.

Browse by category

People

• Andrej Karpathy — AI researcher, creator of the LLM Wiki pattern and AutoResearch.
• Nate Herk — YouTube creator who demonstrated the LLM Wiki in practice.
• Nate B Jones — AI strategy educator; agent primitives, MCP thesis, /loop + OpenBrain.
• Matthew Berman — AI YouTuber focused on open-source models, local inference, and agent tooling.
• Tim Carambat — Founder of AnythingLLM; covers local AI inference and TurboQuant.
• Damian Malliaros — AI creator focused on Claude Code cost optimization and app building.
• Cole Medin — AI educator; Excalidraw diagrams, RAG analysis, agentic engineering.
• Nick Saraev — AI educator; AutoResearch methodology for self-improving Claude Code skills.
• Fireship — Fast-paced developer education YouTube channel.
• Crosstalk Solutions — Home lab creator; built Project Nomad offline AI server.
• Dave Swift — AI tools creator focused on audio and local-first applications.
• Stephen G. Pope — Creator of ThePopeBot; autonomous agent architecture.
• WorldofAI — YouTube creator covering AI tools and agentic IDE launches.
• BMad Code (Brian) — Creator of the BMAD Method; 20-year software vet on agile + AI workflows.
• Joshua Clarke — Cybersecurity educator; local AI for private security workflows.
• Global Science Network — YouTube channel on offline AI, hardware NNs, privacy-first computing.
• Greg Isenberg — Founder/podcaster; startup-ideas community Startup Empire.
• Ras Mic — Engineer at Tempo; non-technical AI infrastructure explainer.
• pixegami — Python LLM tutorial creator; RAG, LangChain, pytest patterns.
• Alex Ziskind — Local LLM benchmarks; vLLM/FP8/Apple Silicon vs Nvidia hardware deep dives.
• The Cyber Mentor (Heath Adams) — Cybersecurity educator; founder of TCM Security Academy.
• Tech With Tim — Python + AI tutorials; concrete walkthroughs of fine-tuning, Ollama, Docker Model Runner.
• Wolfgang’s Channel — Homelab/self-hosting YouTuber; local AI code assistant + power consumption analysis.
• Simon Høiberg — SaaS founder; bootstrapped portfolio, leaving-the-cloud, AI agent monetization playbooks.
• Beyond Fireship — Long-form companion channel to Fireship; from-scratch builder tutorials.
• Zubair Trabzada — No-code AI automation educator; voice agents and n8n workflows.
• Tin Rovic — No-code AI automation creator; n8n workflow recipes.
• Nathan Sebhastian — Code with Nathan; practical dev environment + AI coding tutorials.

Organizations

• Anthropic — US frontier-AI lab; Claude, Claude Code, the MCP standard, the Claude Code architecture leak. PBC structure.
• OpenAI — US frontier-AI lab; ChatGPT, Codex, the Industrial Policy paper. PBC structure.
• Google — Big-tech AI org spanning Gemini (frontier), Gemma 4 (open weights), Firebase Studio, Stitch, Vertex AI, plus Waymo/DeepMind.
• Nous Research — Open-source AI research collective; Hermes Agent + GEPA self-improving loop.
• LangChain — AI framework vendor (LangChain library, LangGraph, Open-SWE); open-source frameworks + paid LangSmith observability.
• Lovable — European no-code AI app builder; afternoon-to-prototype tool referenced across multiple wiki sources.

Models

Tracked separately from tools and frameworks so we can follow each model’s progress, capabilities, and best practices over time.

Frontier (closed-weights):

• ChatGPT — OpenAI’s flagship model; best for ease of use.
• Claude — Anthropic’s model family; best for work and coding.
• Gemini — Google’s model; unique video ingestion; best for search and deep research.
• Grok — xAI’s model; best for real-time Twitter/X research.
• MiniMax M2.7 — Closed-source model claiming to outperform Opus 4.6 at 20x lower cost.

Open-weights:

• Llama — Meta’s open-weights model family; pioneered local LLM inference.
• DeepSeek — Chinese lab open-weights model; stronger than Llama as of 2026.
• Qwen — Alibaba’s open-weights model family; among the top open-source models.
• Gemma 4 — Google’s open-weights model; 31B ranks #3 globally.
• Kimi K2 — Moonshot’s open-weights MoE model; 32B active params, ~10× cheaper than Sonnet at output.
• MiniMax M1 — MiniMax’s open-weights hybrid-attention model; 1M token context, 80K reasoning output.

Tools

• Augment Agent — Agentic IDE; #1 OSS on SWE-bench Verified at 65.4%; large-codebase indexing.
• Firebase Studio — Google’s cloud agentic dev environment (rebrand of Project IDX); free tier.
• GitHub Agent HQ — Microsoft’s evolution of GitHub into a full AI agent orchestration platform.
• Archon OS — Cole Medin’s open-source command center for AI coding agents; unified knowledge + task layer via MCP.
• Sim AI — Open-source visual AI agent workflow builder; n8n alternative with 60+ integrations.
• Supabase — Open-source Postgres + PGVector + auth backend; convergence point for OpenBrain, Archon OS, Sim AI, and AI coding workflows.
• Claude Code — Anthropic’s CLI tool, the engine behind this wiki. Now with /loop and a skills ecosystem.
• Cursor — AI-native code editor; one of the first coding agents.
• Codex — OpenAI’s coding agent harness.
• Devin — Cognition’s autonomous software engineering agent.
• OpenClaw — Fastest-growing open-source agent framework (200k+ stars); security concerns.
• Google Stitch — Voice-to-UI design tool; exports agent-readable design.md.
• Remotion — React framework treating video as code; #1 non-corporate Claude Code skill.
• skills.sh — Open directory for Claude Code agent skills.
• Obsidian — Markdown note-taking app used as the visual frontend.
• Obsidian Web Clipper — Browser extension for clipping articles into raw/.
• Ollama — Local model runner for running open-source LLMs on consumer hardware.
• OpenRouter — Unified API gateway for hundreds of AI models, including free tiers.
• Paperclip — Open-source multi-agent orchestration dashboard; MIT licensed.
• AnythingLLM — All-in-one local AI app for document chat, agents, and automations.
• Open WebUI — Self-hosted ChatGPT-like web interface for local LLMs via Ollama.
• Docker Model Runner — Local LLM runtime built into Docker Desktop; OCI-packaged models, no CUDA needed.
• Continue — Open-source VS Code/JetBrains extension; local Copilot replacement via Ollama or any OpenAI-compliant API.
• n8n — Open-source workflow automation platform; canonical no-code agent runner; convergence point for the wiki’s AI-automation thread.
• Flowith — Multimodal infinite-canvas execution agent; 10M-token context, 30+ native integrations.
• Manus — Closed-source viral general AI agent; the original of the open-source openmanus clone.
• Retell AI — Voice-AI platform for inbound/outbound phone agents; telephony + persona + function calling.
• Zep — Hosted long-term agent memory as a per-user relational graph; n8n integration.
• Void IDE — Fully open-source agentic IDE shipped as a full editor; Agent Mode + Gather Mode.

Open-Source Projects

• llama.cpp — C++ inference engine; the foundation underlying Ollama, AnythingLLM, and most local AI tools.
• Unsloth — Fast, free fine-tuning library for open-weights LLMs; LoRA + GGUF export pipeline that runs on free Colab T4.
• Open-SWE — LangChain’s open-source async AI coding agent on LangGraph; self-hostable, bring-your-own-model, free Jules alternative.
• LangChain (library) — Original Python/JavaScript framework for chaining LLM calls with tools, prompts, retrievers, memory, and document loaders. The first widely-adopted LLM-app framework.
• LangGraph — Stateful multi-step agent framework from LangChain; agents as directed graphs of nodes and edges; underlies Open-SWE.
• Blender MCP — Natural language to 3D via MCP bridge to Blender’s Python API.
• GStack — YC President Gary Tan’s startup methodology as Claude Code skills.
• Hermes Agent — Self-improving agent framework with closed learning loop.
• Superpowers — TDD-enforcing Claude Code plugin; 115k stars.
• Agency — Role-based agent templates for Claude Code (frontend, backend, security, etc.).
• Prompt Fu — Unit testing framework for prompts; acquired by OpenAI.
• Impeccable — Claude Code skill for frontend design; 17 UI commands.
• Open Viking — File-system agent memory DB with tiered loading.
• Mirrorish — Multi-agent prediction engine with simulated social network.
• Heretic — Model guardrail removal via “obliteration” technique.
• Nano Chat — Full LLM pipeline; train your own model for ~$100.
• Project Nomad — Offline knowledge server with local AI, maps, and education.
• VoiceBox — Free local TTS app with voice cloning; open-source ElevenLabs alternative.
• Attractor — StrongDM’s dark factory agent; three markdown specs = the entire agent.
• ThePopeBot — Free open-source OpenClaw alternative; Docker + GitHub Actions + Ollama.
• DeepCode — Open-source multi-agent coding platform from HKUDS; Paper2Code, Text2Web, Text2Backend.
• Bjorn — Open-source Raspberry Pi network pentest tool; defensive audit for self-hosted AI infrastructure.
• vLLM — High-throughput LLM inference engine; saturates Nvidia GPUs with concurrent requests.
• Cline — Open-source autonomous AI coding agent as a VS Code extension; bring-your-own-model.
• OpenCode — SST’s open-source terminal AI coder; 75+ providers, parallel agents, themable UI.
• OpenManus — MetaGPT’s open-source clone of Manus; Python agent loop with no UI.
• Kimi Coder — Llama Coder fork repointed at Kimi K2; web app for prompt-to-SPA generation.
• Docmost — Self-hostable collaborative wiki/docs platform; Notion (docs portion) alternative.
• Plane — Self-hostable open-source project management; Jira/Linear/Notion (PM portion) alternative.
• Strapi — Self-hostable headless CMS; Contentful/Sanity alternative.
• NocoDB — Self-hostable no-code database; Airtable alternative.
• Grafana — Open-source observability/analytics dashboards; PostHog/Amplitude alternative when paired with Postgres.

Concepts

• LLM Wiki Pattern — Method for building LLM-maintained personal knowledge bases in markdown.
• Agentic Harness Primitives — 12-primitive framework for production-grade agent systems.
• Open-Source Model Integration — Running Claude Code with local or third-party models to reduce cost.
• Multi-Agent Orchestration — Coordinating multiple AI agents with distinct roles toward shared goals.
• TurboQuant — KV cache optimization enabling 4x more context window on the same consumer hardware.
• loop — Anthropic’s proactive agent scheduling command for Claude Code.
• OpenBrain — SQL database + MCP server pattern for persistent agent memory.
• AutoResearch and Evals — Self-improving methodology with objective metrics and binary evals.
• Four Prompting Disciplines — Prompt craft → context engineering → intent engineering → specification engineering.
• Five Levels of AI Coding — L0 spicy autocomplete → L5 dark factory. Dan Shapiro’s framework + StrongDM case study.
• Frontier Operations — Five persistent skills for working at the expanding AI/human boundary.
• BMAD Method — Agile AI-Driven Development; six AI personas (BA → PM → Architect → PO → SM → Dev).
• AI Coding Workflow — Cole Medin’s PLANNING.md + TASK.md + global rules + MCP servers pattern.
• Model Context Protocol (MCP) — Anthropic’s standard for connecting LLMs to tools, services, and data.
• Meta-Prompting — Wrapping sloppy requests in a structured instruction-set; power steering for GPT-5.
• Fine-Tuning (LLMs) — LoRA + GGUF + Ollama pipeline for specializing open-weights models on local hardware.
• FP8 Quantization — Floating-point 8-bit precision; native to Nvidia Blackwell tensor cores; the throughput sweet spot for vLLM.
• Context Engineering — Karpathy/Lütke’s successor to prompt engineering; curating all the information an LLM needs to plausibly solve a task.
• Claude Code Sub-Agents — Anthropic’s first-party multi-agent primitive: isolated context, tool whitelists, parallel execution.

Analyses

• RAG vs Wiki — Comparison of semantic search RAG vs the LLM Wiki approach.
• SaaS Death Spiral — Why $1T in SaaS market cap evaporated in early 2026; seven AI developments and the per-seat pricing axis.
• Self-Hostable Alternatives — Running map of commercial dev tools and credible open-source self-hostable replacements captured in this wiki.

Personal Hardware Benchmarks

• Benchmark Index — Master ranking of all 16 personal rigs by AI Capability Index.
• Benchmark Methodology — How the AI Capability Index (ACI) is calculated.
• Gemma 4 VRAM Requirements — GPU + Mac lookup tables; quantization breakdown.

Business

• AI Automation Workflows — The 5 highest-value AI automations to sell to businesses, plus the clogged pipe sales framework.
• AI Professional Interface — Replace the broken hiring pipeline with an AI-powered personal interface. Five components, attention economics, bidirectional fit assessment.

Sources

• AI 2027 — Forecasting article used as a demo ingest source.
• Industrial Policy for the Intelligence Age — OpenAI’s first major public policy paper (April 2026); 21 proposals for the transition to “superintelligence.”
• Karpathy 10x’d Claude Code — Nate Herk demonstrates the LLM Wiki in practice.
• 12 Agent Primitives — Nate B Jones analyzes the Claude Code leak.
• Ollama + Claude Code — Running Claude Code with local and free cloud models.
• Google Drops Gemma 4 — Matthew Berman covers Gemma 4 benchmarks.
• Claude Code + Paperclip — Multi-agent orchestration with Paperclip.
• TurboQuant — Tim Carambat covers TurboQuant’s impact on local models.
• MiniMax M2.7 + Claude Code — Building a ClickUp clone at 95% lower cost.
• 500 AI Workflows — Nate Herk’s 5 highest-value automation workflows.
• Design Meeting → Markdown File — Google Stitch, Remotion, Blender MCP.
• What Is Llama.cpp? — GGUF format, quantization, local server mode.
• Open-Source AI Projects — GStack, Hermes Agent, Superpowers, Paperclip.
• Every AI Model Explained — Frontier models, open-source models, coding agents.
• 7 Open-Source AI Tools — Agency, Prompt Fu, Impeccable, Open Viking, and more.
• loop + OpenBrain — Three agent primitives; OpenClaw without the risk.
• Project Nomad — Offline knowledge server with local AI.
• AutoResearch for Skills — Self-improving Claude Code skills with evals.
• VoiceBox TTS — Free open-source local text-to-speech.
• Prompting Split Into 4 Skills — The four-discipline prompting framework for 2026.
• Excalidraw Diagrams — Visual validation loop for Claude Code diagram generation.
• RAG for Code vs Knowledge — Why coding tools abandoned RAG; when semantic search still matters.
• OpenBrain Architecture — The $0.10/month agent memory system in full detail.
• 5 Levels of AI Coding — Dark factory, StrongDM, J-curve, junior pipeline collapse.
• FREE OpenClaw — ThePopeBot architecture: Docker + GitHub Actions + Ollama.
• E2E Testing Command — Self-healing AI coding workflow with browser automation.
• Frontier Operations — The expanding bubble; five persistent skills.
• AI Professional Interface — Replace hiring pipeline with AI-powered personal interface.
• Augment Agent: RIP Cursor — WorldofAI walkthrough of Augment Agent’s SWE-bench launch.
• Better Than Vibe Coding — BMad Code introduces the BMAD Method (six Agile AI personas).
• Local AI for Cybersec — Joshua Clarke on Ollama + Open WebUI for private security writing.
• Code 100x Faster with AI — Cole Medin’s full AI coding process; one-shots a Supabase MCP server.
• Gemma 4 VRAM Requirements — Reference article: GPU + Mac sizing for every Gemma 4 model.
• Firebase Studio Walkthrough — WorldofAI on Google’s free agentic dev environment.
• Dolphin Llama 3 Offline — Running an uncensored Llama variant from a USB drive via Ollama + AnythingLLM.
• MCP Clearly Explained — Ras Mic’s non-technical walkthrough of MCP on the Greg Isenberg podcast.
• Python RAG Tutorial (Local LLMs) — pixegami builds a PDF RAG app with chunk IDs and LLM-as-judge tests.
• ChatGPT-5 Prompting is Too Hard — Nate B Jones on meta-prompting GPT-5; the speedboat with a big rudder.
• 7 Self-Evolving — WorldofAI walkthrough of GEPA, install, Obsidian skill demo.
• How AI is Breaking SaaS — Fireship’s $1T market-cap thesis; seven Feb 2026 developments.
• Docker Model Runner Walkthrough — WorldofAI on the Docker-native local LLM runtime.
• Planning in Claude Code Just Got a Huge Upgrade — Nate Herk on Claude Code’s new Ultra Plan: cloud Opus 4.6 + multi-agent exploration.
• Archon OS Walkthrough — WorldofAI on Cole Medin’s AI coding command center.
• DeepCode Walkthrough — WorldofAI on HKUDS’s multi-agent coding platform with Paper2Code.
• Sim AI Walkthrough — WorldofAI on the open-source n8n alternative for AI agent workflows.
• THIS is the REAL DEAL for local LLMs — Alex Ziskind on vLLM + FP8 + RTX PRO 6000; 5,800 tok/s sustained.
• Meet Bjorn, the Easy to Build Hacking Tool — TCM walkthrough of building a Raspberry Pi network pentest device; defensive audit angle for self-hosters.
• Industrial Policy for the Intelligence Age — OpenAI’s first major public policy paper; 21 proposals + heavy interested-party caveats.
• EASIEST Way to Fine-Tune a LLM and Use It With Ollama — Tech With Tim: Unsloth + LoRA + GGUF + Ollama Modelfile workflow.
• Host Your Own AI Code Assistant with Docker, Ollama and Continue! — Wolfgang’s Channel: 7900XTX gaming PC vs MiniPC for local Copilot; “GPU is the cliff.”
• Open-SWE: Opensource Jules! — WorldofAI on LangChain’s open-source async coding agent built on LangGraph.
• 5 (Real) AI Agent Business Ideas For 2025 — Simon Høiberg’s five concrete AI monetization paths.
• Replace Cloud Tools With Self-Hostable Alternatives — Six-tool replacement stack saving $10K/yr.
• Claude Code Sub-Agents — WorldofAI on Anthropic’s first-class sub-agents in Claude Code.
• Cline v3.18 HUGE UPDATE — Gemini CLI provider, Claude Code as backend, 35 new MCP servers.
• Context Engineering: The End of Vibe Coding — Walkthrough of Cole Medin’s PRP template.
• Flowith Neo: Infinite Agent — Multimodal infinite-canvas execution agent.
• I built a DeepSeek R1 powered VS Code extension — From-scratch VS Code extension wired to Ollama + DeepSeek R1.
• Kimi Coder — Kimi K2 model + Llama Coder fork for prompt-to-SPA generation.
• MiniMax Agent (M1) — Open-weights hybrid-attention model + the hosted general agent.
• Docker MCP Toolkit — Containerized MCP install + secrets vault inside Docker Desktop.
• OpenCode — SST’s terminal AI coder with 75+ provider support.
• OpenManus — MetaGPT’s open-source Manus clone.
• AI Voice Agent (Retell + n8n) — Inbound dental receptionist no-code build.
• YouTube → SEO Blog (n8n) — Content-repurposing pipeline with n8n + OpenRouter + WordPress.
• n8n + Zep memory — Knowledge-graph long-term memory with cost-trap fix.
• VSCode + Cline + Continue — Canonical free Cursor-replacement setup.
• Void IDE Update — Fully open-source agentic IDE with Agent Mode + Gather Mode.

Contribute or build your own

This wiki is open source at github.com/jessefreeman/ai-for-dev. If you spot something wrong or want to improve a page, open a pull request — the site rebuilds automatically as content is added.

Want to build your own? Clone the repo, delete wiki/ and raw/, then follow the instructions in the README to start from scratch with your own content.