Neural Lattice · Toolbox

The Toolbox

Every link here survived the cut — nothing that's just SEO bait with a gradient logo. Curated collections for builders, researchers, and basement tinkerers who want signal, not noise.

The Starter Loadout

If you only bookmark six things this year, make it these. No notes app required — these cover 80% of what you'll ever need to look up.

Hugging FaceModels, datasets, Spaces — the town square of open ML. If it's not on HF, it probably doesn't exist yet. Papers With CodeResearch papers + code implementations + benchmark leaderboards in one search. The fastest way from "I heard about a paper" to "I ran it myself." Claude.aiAnthropic's chat — projects, artifacts, long context. Still the best spot for deep reasoning work and code analysis. ChatGPTOpenAI's consumer chat with GPT-4o, voice, custom GPTs, and Canva Code 2.0 integration. The Swiss Army knife. Google AI StudioGemini API playground with a free tier that's genuinely usable for production prototyping. Long context wins. GitHub CopilotStill the coding assistant benchmark. Agent mode, multi-file edits, and terminal integration in 2026 — ambient developer pairing.

Build Your Own Basement Stack

Privacy-first, no cloud lock-in, and you own the hardware. This is the stack running on the Ayra ix NAS — and you can replicate it with a weekend and a spare GPU.

OllamaThe default local inference layer. Pull models, serve an OpenAI-compatible API, manage quantizations. If you self-host, this is your first install. Open WebUISelf-hosted ChatGPT-style frontend for Ollama. Multi-user, RAG pipelines, MCP tool routing — the dashboard for your basement brain. llama.cppThe engine under Ollama and LM Studio. C++ inference for GGUF models — what makes local LLMs actually fast enough to use daily. LM StudioDesktop GUI for downloading and running GGUF models offline. Best entry point if you want local AI without touching a terminal. LiteLLMUnified API gateway for 100+ providers. Swap between Claude, GPT, and your local Ollama models without changing a line of app code. QdrantProduction-grade vector database in Rust. The search backbone when Chroma isn't enough but Pinecone is too expensive.

Benchmarks That Don't Lie

Model rankings without the marketing spin. These are the leaderboards that actually tell you which model wins for your use case — not just which lab has the best PR team.

LMSYS Chatbot ArenaBlind head-to-head model voting from real humans. The closest thing to a live preference leaderboard — GPT-5.6 Sol leapfrogged everyone here in Jul 2026. Artificial AnalysisIndependent speed, price, and quality comparisons across hosted providers. Answers "which model is fastest for my budget" without the sandbagging. Epoch AIResearch on compute trends, training costs, and scaling laws. The data behind every "AI is moving too fast" headline you've read this year. LiveCodeBenchCoding benchmark that pulls fresh problems to prevent contamination. The only leaderboard where "trained on the test set" isn't a cheating factor. Open LLM LeaderboardHF's standardized benchmark suite for open-weight models. Runs the same tests on every submission — useful for comparing Llama, Qwen, Mistral variants.

Documentation That Doesn't Suck

These docs teams actually wrote for developers, not compliance reviewers. Quick-reference bookmark set for anyone shipping AI features.

Anthropic DocsClaude API reference, prompt guides, tool-use patterns, and the best explanation of extended thinking you'll find anywhere. OpenAI Platform DocsGPT-4o, vision, audio, assistants API, and structured outputs. Everything you need to build on the platform that started the API wars. Ollama API DocsLocal OpenAI-compatible endpoints — essential reading for the self-hosted stack. One curl command to verify your setup is working. MCP Protocol DocsThe standard for connecting AI models to external tools. If you're building agents in 2026, this is the integration pattern to learn.

Worth Your Inbox Space

Newsletters that earn their place in your morning read. Zero PR rewrites, actual analysis, and they don't all say the same thing.

The Rundown AIDaily AI news digest for the broad picture — 500k+ subscribers for a reason. Best for "what happened yesterday" in 5 minutes. Ben's BitesCurated daily AI news with a builder's lens. Less corporate, more opinion — reads like a group chat with someone who actually ships. Import AIJack Clark's newsletter on the research edge. Longer form, deeper technical dives — think of it as the arXiv digest for busy builders. InterconnectsNathan Lambert's take on open-source AI, RLHF, and the politics of open-weight models. The best source for understanding the open vs. closed debate.

Rabbit Holes Worth Falling Into

Not essential for shipping — but if you have a Sunday afternoon and want to remember why AI is actually interesting, start here.

Common CrawlPetabytes of web crawl data — the raw ingredient behind most pre-training corpora. Fun to search, humbling to contemplate the scale. FuturepediaLargest AI tools directory — filter by use case, pricing, and category. Good for "I need a tool that does X" exploration sessions. There's An AI For ThatType any task and get a list of AI tools that do it. Surprisingly addictive — "there's an AI for vacuuming my inbox?" Yes. NotebookLMGoogle's AI notebook that turns your documents into podcast discussions. The "audio overview" feature still feels like magic a year later. ElicitAI research assistant that finds papers and extracts findings. Like having a grad student who only sleeps when you do.

AI IQ Check

Question 1 of 5