A living place where step-by-step recipes, the repos and models taking off on GitHub and Hugging Face, and real cases that deliver value all live together. Open a recipe, it asks about your business, you answer, and it walks you through one step at a time. Like a Thermomix, but for AI.
El cliente escribe por tercera vez y por tercera vez le preguntan el número de pedido. Con una memoria por cliente conectada a n8n, el asistente revisa lo que ya se habló antes de redactar la respuesta y guarda lo nuevo al terminar. Tú apruebas antes de enviar.
Media jornada para armar el flujo; después funciona solo y lo revisas una vez al mes
La planilla se llama COSTOS_FINAL_final_v3.xlsx y la hizo alguien que ya no trabaja aquí. Nadie sabe qué hace la celda F47, pero suma. Una skill abierta de Anthropic revisa las fórmulas celda por celda y te entrega una tabla con cada error, su gravedad y cómo arreglarlo, sin tocar nada.
Una hora: diez minutos de instalación y el resto revisando lo que encontró
Un modelo abierto de voz a texto pasa cada nota de voz a texto y la IA la convierte en una ficha corta: quién es, qué pide, qué tan urgente es y qué falta confirmar. Se prueba gratis en el navegador, sin instalar nada.
Media hora para probarlo con tus audios; dos horas si lo dejas instalado en un computador de la oficina
Con la lista de facturas impagas y el historial de pagos, la IA separa a los buenos pagadores de los que siempre se atrasan y redacta un correo por cliente, amable o firme según su historial. Tú revisas y apruebas; nada sale sin tu visto bueno.
Una hora la primera vez; quince minutos cada semana
Una automatización lee cada lunes las páginas de producto de tus competidores, saca precio y stock como datos ordenados y los deja en una planilla junto a los tuyos. La IA te resume qué cambió y dónde estás caro.
Tres horas para armarlo; después corre solo y lo revisas en diez minutos cada lunes
A chat that answers questions from customers or your team using only your manuals, policies and catalogs, and says "I don't know" when the answer is not there.
Every comment, review or community message goes through a filter that detects insults, spam, personal data and off-topic content, and only what passes is published on its own.
One terminal command shows you exactly the HTML that the OpenAI, Anthropic and Perplexity bots receive. If it is empty, your company does not exist for AI.
Most business problems are solved better with a fixed flow of steps than with an autonomous agent. This recipe helps you decide which one you need before spending a peso.
Role, context, task, output format and examples. Prompts that fail almost always leave out two of the five. This recipe is the template we use in all the others.
Every month you compare what you spent on each line against what was expected and end up with a short list: only what broke the pattern, with its likely explanation.
One hour the first time, fifteen minutes every month
Every week you receive a data-backed proposal for what to feature on your online store's homepage: which products, which promotion and which copy, based on your sales and your calendar. You approve and publish.
One afternoon to set it up, ten minutes every Monday
You upload a long file (reports, records, emails, contracts) and receive a structured summary of what matters for your decision, each data point with the page it came from, and a list of what is missing or contradictory.
Two hours to build the template, minutes per case file
Before signing a lease, supplier or service contract, you receive a plain-language summary of what binds you, what protects you and which clauses are worth negotiating, with the exact citation for each one.
A Project with your manuals, price lists and policies uploaded once: every conversation starts knowing how your business works and answers from your documents, not with generalities.
MCP is the standard plug between AI assistants and external tools. With this recipe you connect the first one, understand what it can and cannot do, and are ready to connect your own.
Half an hour, including the first conversation with the connected tool
Every week you reply to the new reviews on your Google Business Profile, good and bad, with texts that sound like you, thank people for specifics and resolve complaints without arguing. You review and paste.
An open source CRM with AI agents that reply, ask the qualification questions and leave every conversation sorted in a pipeline, running on your own server. No per-seat or per-message fees.
Three days: one to install, two to tune the replies with real conversations
You upload manuals, contracts, procedures and spreadsheets; your team asks in plain language and gets the answer with the exact citation. With connectors to Notion, Telegram or Slack so it lives where they already work.
A program looks at your computer, tells you which AI models it can run well and how fast, downloads them and gets them ready. For data that cannot leave the company and for working offline.
Forty-five minutes, most of it waiting for the model download
The agent uses your own browser, already signed in to your bank, your supplier portal or your sales dashboard, in a separate window, and hands control back to you when there is a captcha or a confirmation. You keep working.
With your sales history in a Google Sheets column and a forecasting model trained by Google, you get next month's projection by product or branch, to buy and staff better.
Before paying for a tool or asking someone to build it, you test for free in the browser the open model that would solve it: generating images, transcribing audio, classifying text, translating. In ten minutes you know whether it works.
With ten seconds of your voice, a free local program generates the voiceovers for your product videos, announcements and tutorials. You type the script, your voice comes out. Nothing is uploaded to the internet.
You give it the long video with its subtitles; the tool finds the moments worth keeping, cuts them, adds a title and exports them in the format for TikTok, Reels and Shorts. You choose which ones to publish.
Forty-five minutes for the first video, then fifteen
Before sending any text a model helped you write, you run it through a skill that removes the machine tics (template phrases, forced lists, brochure enthusiasm) without changing what it says.
Your AI agent learns to read and search X, Reddit, YouTube, GitHub and web pages. You ask it "what are people saying about [my category] this week?" and it brings back a summary with links, without paying for a social listening tool.
With an open source SEO tool that includes AI engine visibility, you find out which questions the assistants mention you for, which ones they do not, and which keywords are worth working on. And you let your agent do the analysis.
A simple rule for deciding which model to use for each job: the most capable one for what is hard and important, the fastest and cheapest for what is repetitive. With a table you build in twenty minutes and use all year.
Twenty minutes
radar // github + hugging face
On the radar today: GitHub and Hugging Face
The all-time best and what is moving most in the last 15 days, 11 and 11 per source, no repeats. Refreshed every morning.
reading the radar…
What a recipe is
A real case, published by a company, a repository or a verifiable outlet, turned into instructions you can follow today. Plus a guided mode: it asks about your business, builds your prompt and walks you through one step at a time.
How it is verified
Every recipe carries its original source with a link and a verification date. Figures are quoted exactly as the source publishes them; if the source gives no numbers, neither does the recipe. Never two recipes about the same thing.