Local AI & Automation Editor · since 2019
I'm US-based, I run more models at home than I'll admit to, and I've quantized more than I've finished reading about. I write about running AI on your own hardware and, lately, about what it costs a company to do the same — tokens per day, GPUs per month, and the GDPR questions nobody's sales deck answers.
One person on one machine, or a team hitting one server? That single question decides it. Then licence, GPU and Mac support, and the API your app talks to, all read from each project's own docs.
Tensor · 2d ago · 9 min
Olympiad golds, Erdős problems, a disproved 87-year-old conjecture and a better Riemann bound. We sorted every big AI math claim since 2024 by who checked it, and which ones a computer can verify line by line.
Tensor · 3d ago · 11 min
A hook is a small script Claude Code runs for you at fixed moments, such as before a tool call or when Claude finishes. Here is every piece in Anthropic's docs that matters: events, matchers, exit codes, JSON, where the file lives, and the security warning.
Tensor · 3d ago · 11 min
Article 50 has applied since 2 August 2026. If your chatbot or voice agent talks to people in the EU, the first duty is simple: tell them it is an AI. Here is the full checklist, who owes what, and the fine tier.
Tensor · 4d ago · 9 min
Plain OCR costs $1.50 per 1,000 pages on all three clouds. Invoices cost $10 on two and $100 on the third. We priced text, forms with tables and invoices per page, then checked the LLM-vision shortcut.
Tensor · 4d ago · 11 min
ChatGPT Business and Claude Team are both $20 a seat on annual billing. Microsoft's Copilot add-on looks cheaper at $18 but needs a Microsoft 365 license under it. We priced four AI seat plans at 10, 25 and 50 users and read each vendor's data terms.
Tensor · 4d ago · 9 min
Vapi advertises $0.05 a minute and Retell $0.07. Neither is what a phone call costs. We split the all-in minute into platform fee, speech, LLM and phone line for seven voice-agent routes, then priced a five-minute US call at 1,000 and 10,000 calls a month.
Tensor · 4d ago · 11 min
Renting a GPU for an open model looks cheaper than paying per token until you price the same model on the cheapest API. We did it for four open models, from Llama 3.1 8B to Llama 3.3 70B, with live GPU rates from our own tracker and list prices read on 2 October 2026. Against the same model's API, one rented GPU almost never wins. Against a frontier model, it wins from about 2 to 6 million tokens a day.
Tensor · 4d ago · 12 min
Every AI support platform bills a different thing: an outcome, a verified resolution, a conversation, a session, an action or a credit. We priced the same support queue on all of them at 1,000, 5,000 and 20,000 chats a month and 40% to 80% resolved, and quoted what each vendor counts as a resolution in its own words.
Tensor · Sep 27 · 13 min
Keeping prompts inside the EU usually adds 10% to a frontier model's bill, 20% for GPT-6 on Azure, and nothing on Mistral. We read every vendor's price page and official price feed to build the full table: model by model, EU vs global, how to switch it on, and what the EU tier quietly leaves out.
Tensor · Sep 27 · 12 min
Your company wants AI on internal documents without the prompts leaving the EU. We priced every route on the vendors' own pages: OpenAI's and Azure's EU tiers, Claude through AWS and Google, Mistral, French and German open-model APIs, and a GPU of your own. The EU premium is 0% to 75%, and for most companies the cheapest private option isn't a server.
Tensor · Sep 27 · 14 min
You've got a 24GB card and three excellent dense models that fit it. Here's how Mistral Small 24B, Qwen3.6 27B and Gemma 4 31B differ on VRAM, context and modality — and a clear pick for each kind of user.
Tensor · Sep 26 · 7 min
Qwen2.5-Coder-32B still runs a serious private coding assistant on a single 24GB card, with a 32K native context (128K with YaRN). The VRAM math, the tight-but-real fit, and how it stands against the newer Qwen3-Coder.
Tensor · Sep 26 · 7 min
Microsoft's Phi-4 is a compact 14B that runs on a mid-range card — but at 16K context and a late-2024 release, it shows its age next to newer 14B-class models. The VRAM math and an honest verdict.
Tensor · Sep 26 · 7 min
Qwen3.6's dense 27B is the no-fuss 24GB-card option — every parameter fires, nothing to offload, a real 256K context. The VRAM math, the GPU, and how it differs from the 35B-A3B MoE.
Tensor · Sep 26 · 7 min
Qwen3-Next 80B-A3B activates only 3B parameters per token — so it's fast. But 80B of weights still have to live somewhere. The honest VRAM math, the offload trick, and what it really takes.
Tensor · Sep 26 · 7 min
Gemma 4's mixture-of-experts model holds about 26B parameters but activates only 3.8B — so it's quick, and it fits a 16GB card at Q4. The VRAM math, the GPU, and the run command.
Tensor · Sep 26 · 7 min
Gemma 4 31B is the big dense Gemma — the one that asks for a 24GB card and rewards it. The VRAM per quant, the GPU that fits, and how to run it without pulling the wrong tag.
Tensor · Sep 26 · 7 min
Qwen3.6's default is a mixture-of-experts model: 35B on disk, 3B active per token. That makes it fast — but the VRAM math has a twist worth understanding before you buy a card.
Tensor · Sep 26 · 7 min
Type ollama run qwen3.5 and this is the model you get — the 9B. What it is, how little VRAM it needs, and the cheapest card that runs it well.
Tensor · Sep 26 · 7 min
Google's Gemma 4 12B is the multimodal open model that fits a mid-range card — if you pull the right Ollama tag. The honest VRAM math, the cheapest GPU that runs it, and how to start.
Tensor · Sep 26 · 7 min