The questions people ask before they trust a memory with their AI. Short answers, no marketing.
A memory you own that every AI you use can read, and a team of helpers (or positions, for a business) that work from that memory — so you say things once, and nothing is lost when you switch from ChatGPT to Claude to Grok.
Mostly, and here is the honest version. The connector gives the app a set of tools: read my memory, save a fact, save a checkpoint. Cortesso tells the app to use them at the start and end of every conversation. Claude follows that on its own. ChatGPT and others follow it far more reliably when you paste one paragraph into the project's instructions (Cortesso gives you the paragraph, with a Copy button, on each position's page). What the connector does not do is copy every chat you have in those apps word for word — no vendor lets a connector read your chat history. For that, export your history once from ChatGPT or Claude and drop the file into Memory; from then on the apps save what matters as they go.
With Claude, no. With ChatGPT, one paragraph in the project (or your custom instructions) makes it dependable; it takes ten seconds and Cortesso hands it to you. Inside Cortesso itself — the Office, helpers, the CFO — nothing to write: they read your memory every time.
You can. Once a plan or a key is set up, the Office is one place to ask any helper or position, and each one runs on the engine you chose — Claude for one, ChatGPT for another, Grok for a third — with the same memory behind them. "Ask everyone" sends one question to all of them side by side. You can still use the apps you already pay for; Cortesso connects to them too. Most people end up doing everyday work in Cortesso and keeping the apps for what they are uniquely good at.
Three minutes: sign in, drop your ChatGPT export (or type three things about yourself), ask a helper a question it could only answer if it knew you. Then open Claude with the connector and ask the same thing. That moment — two different AIs, one memory — is the product.
The same thing with two names. A title, a sentence about what it is for, a few rules, and which AI it runs on. "Planner", "Buyer" and "Writer" for you; "CFO", "Head Designer", "Ops" for a business. Hire one in a click; edit its rules any time; let it go.
It keeps books. Import your bank and card statements; the CFO computes the P&L by month and the cash per account from the numbers, reconciles them with what the bank reported, compares them with your budget, and writes the monthly close on the 1st. When it gives you a figure it says "the ledger" or the file name. A chatbot with a finance prompt guesses; the CFO adds up.
No. Cortesso never trains on your data or sells it. When a helper runs on Claude or ChatGPT, it goes through the vendors' API, whose terms exclude API traffic from training. The Trust page has the table, vendor by vendor.
You, and the apps and helpers you connected — each only within what you granted, and each revocable in one click. Content is encrypted at rest with a key made for your account. Books stores dates, amounts and categories in clear so they can be added up; the Trust page says exactly that and nothing more.
Yes. A team is a second house with its own shared memory, positions, board, Books and weekly brief; you invite people by email with a title ("Ops lead") and they join as human positions. Their personal memory stays theirs; the team only sees what they do inside it. Runs they start use their own engine key first, then yours or your plan. Every Monday the brief goes to everyone in the team by email. Team is per seat: one seat per person, each with their own monthly allowance; add or remove seats any time from the billing portal.
Yes: a photo of a receipt becomes one transaction in Books (date, vendor, total), and a photo added to a project becomes a file every position can read. The reading is done by your own engine (Claude, GPT or Gemini key, or your plan), so it costs a few hundred tokens and shows under Tokens — and you should check the figure against the paper.
Eight weeks, from the ledger: what repeats (rent, payroll, subscriptions, a customer who pays monthly), plus what everything else has netted lately, against the balances it knows. It says what it assumes, and it needs three occurrences before it calls something recurring.
Today → Numbers → "Every run, by project". Every run with who asked, the position, the project, the ask, tokens in and out and the cost, filtered any way you like, and a CSV to download.
Always. Export gives you everything as one readable zip; Delete everything removes it all, including your encryption key, from Settings, with no email and no waiting. Never capped, on any plan.
Free gives you ten answers a month on Cortesso's own accounts — Claude, ChatGPT, Gemini and Grok — with no API key and no card, plus your memory and the connector that carries your work between AIs. Cortesso is $10 a person a month and is everything else: positions with job descriptions and rules, the board, projects, files, Books, the shared company memory, the people you invite (a seat each), the Monday brief, Ask everyone and the Roundtable. It runs on your own API keys, so the engine bill is yours at the vendor's price and never marked up; a key from Anthropic, OpenAI, Google or xAI takes two minutes and is free to create.
Because the price is for Cortesso, not for tokens. A plan that included engines would have to be priced by how much you use them, which is the one thing you should not have to think about. Your key has no allowance from us, only the daily caps you set on it in Settings, and the dashboard shows what each answer cost. The free tier is the exception: ten answers a month on Cortesso's own accounts, so you can see it work before you fetch anything.
Stripe, on behalf of Combat Textiles LLC, which operates Cortesso until Cortesso Inc. is formed; the statement reads COMBAT TEXTILES CORTESSO. Cortesso never sees the card.
Connected through their own apps: Claude (app and Claude Code), ChatGPT, Gemini CLI, and anything else that speaks MCP. Running inside Cortesso with your key, on any position: Claude, GPT, Grok, Gemini, Perplexity, Mistral, DeepSeek, Groq (Llama, GPT-OSS, Qwen) and OpenRouter (one key, hundreds of models). A team can mix them: a CFO on Claude, a researcher on Perplexity, a drafter on DeepSeek — and in the Office's Roundtable they answer the same question and then each other.
The dashboard is a web page that works on a phone; sign in with your email and a six-digit code. A native app is not planned until the web one is finished.
Positions keep answering. When xAI retired a search feature, Grok answers went through a different door the same day and still answered without the web in the meantime. Your memory is yours in a portable format either way.
Help has the three-minute start and the "stuck?" list. Otherwise hello@cortesso.com; say what you asked and what came back. We never need your key or your files to help.