How it works
One sentence, and it’s filed.
Giraffe is where the conclusions of your AI work end up. It keeps the decisions, what’s true now, the open questions and the notes, organized by project, so you can find what you concluded without reopening the conversation.
What you actually say
There’s no syntax and nothing to memorize. Once the connection is on, these are ordinary sentences, and they are the whole interface:
“Save what we concluded to my Acme project.”
“What are the open questions on the kitchen remodel?”
“Update the contract. It came in at $48k.”
“What did we decide about the flooring, and when?”
Your assistant can create the project too, if you don’t have one by that name yet.
The four steps
Work normally
You don’t change how you use AI. Research a decision, compare options, draft a plan, troubleshoot a problem, in whatever assistant you already reach for. The one thing you do first is add Giraffe in your assistant’s own settings, on a computer, and turn it on in the chat you want it in. That part can’t be done from a phone, but once it’s done, Claude works on your phone as normal, while ChatGPT’s custom apps stay on the web. That’s the whole setup, and the guides below are the exact clicks.
Save what matters
When you reach something worth keeping, like a decision, a set of facts that are true right now, or the questions still open, you ask your assistant to save it to a project, in the words above. Your assistant sorts it on the way in, into headings Giraffe already provides: the number goes to Current, the call you made goes to Decisions, the thing you couldn’t settle goes to Open questions. Giraffe keeps the conclusion, in plain language, not the back-and-forth that produced it. You choose the moment, and you choose what goes in.
Open it when you need it
Days or years later, when you need what you worked out, you open the project and read it. What’s true now, the decisions you made, what’s still open, what changed recently and which assistant changed it, all on one page. No scrolling a chat you can barely find again. If you can’t remember which project it’s in, one search box covers all of them.
Continue only if you want to
Often, reading is enough. When you do want to keep going, the project isn’t trapped in the assistant that made it. The app calls it Copy for AI: one button puts the whole project on your clipboard, ready to paste wherever you’re working next, or you can just ask the assistant you’ve connected to read the project and pick up from there. Either way you start from where you actually landed instead of explaining it all again. Using more than one AI is a convenience here, not the point.
Which AIs it works with
Claude works today, for reading and saving. You add Giraffe in Claude’s own settings, on a computer, and then turn it on in the chat you want it in: that second step is the one people skip. It works on every Claude plan: Free allows one custom connector at a time, and on Team and Enterprise an owner has to add it for the organization before anyone else can connect. The Claude setup guide is the exact clicks.
ChatGPT needs a paid plan, on the web, with Developer mode turned on. What we tested: on 12 August 2026 a ChatGPT Plus account, on the web with Developer mode on, read a project and saved a decision to it. Two more things before you try it. It doesn’t work inside a ChatGPT Project, so use an ordinary chat. And by default ChatGPT asks you to approve each save, so nothing leaves the chat without you saying yes.
That was one account on one day, not a promise about yours. OpenAI’s help center and its own developer documentation disagree about which paid plans can save, and in a Business or Enterprise workspace your administrator can restrict it, so the ChatGPT setup guide gives you only what we tested ourselves, with the date we tested it.
Gemini works too, through Gemini Spark rather than an ordinary Gemini chat, a newer, still-rolling-out Google feature. What we tested: on 18 August 2026 a Google account connected in Spark’s Custom apps, read a project and created a Document through a Spark task. Setup happens on the web; once connected, Spark works on mobile too. Whichever assistant you use, the projects stay in Giraffe, so what you concluded is never stuck inside one of them. And you don’t need any of this to start: you can create a project in the browser and write what’s true now, the decisions, the open questions and the notes by hand.
Already using ChatGPT Projects or Claude’s memory? Keep them. They’re built to make the next answer better. Giraffe is built so you can read the record without asking anything. Different jobs, and you can have both: why the two kinds of memory are not the same.
Who it’s for, and who it isn’t
If your work with an assistant ends in something you’ll need again, this is for you. A renewal you have to walk into a room with. A remodel whose budget keeps moving. A decision you don’t want to take twice. A topic you keep coming back to.
If you mostly ask one-off questions and never look back, you don’t need Giraffe, and we’d rather say so than sell you a fifth tab. It works whether you use one assistant or several. Giraffe isn’t a bridge between them, it’s where the work lands once you’ve done it, whichever one you used. See the kinds of work Giraffe fits for four projects written out in full.
What a project keeps
A project isn’t another saved chat. It’s what you’d want to read if you opened it cold a month from now: the conclusions, not the conversation, kept up to date as you add to it. These are the app’s own headings for the things you’ll use most, so what you read here is what’s waiting for you when you sign up.
- Current
- What’s true right now: the figures, dates and status you keep coming back to.
- Decisions
- The calls you made, and what they replaced, so you don’t re-argue them later.
- Open questions
- What’s still unresolved, gathered in one place instead of lost in a thread.
- Notes
- What you noticed and didn’t want to lose. Quartz lead times jump after March.
- Sources
- Which assistant saved each update, and whatever it recorded about the conversation it came from.
- Recent changes
- What moved over time, and where it moved from: the $56k that became $48k.
Finding it again
Everything you save is searchable from one box. ⌘K on a Mac, Ctrl-K everywhere else, from any page in the app. There’s a Search button in the sidebar too, for the times your hands aren’t already on the keyboard. It searches every project you have at once, filed or archived, whether an assistant wrote it or you typed it in yourself. There’s no folder to open first and no scope to set. If you want a narrower answer, you type another word.
You aren’t searching a transcript for the sentence you typed at the time. You’re searching short labeled things, like a fact called Budget, a decision about the flooring, or a question nobody has answered yet, so a result is one line long and says what kind of thing it is, which project it lives in and when it last moved. You can tell which one you wanted without opening any of them.
It matches the words you type, not what you meant by them. Plurals and tenses sort themselves out and part of a word will find the whole of it, but there’s no rephrasing and no guessing at what you were getting at. Use fewer words rather than more. One distinctive one usually does it, because you’re matching short labels rather than paragraphs.
The filters along the top are the kinds of thing you matched, each carrying its count: All, Current, Decisions, Questions, Notes, Projects, Sources. Only the kinds you actually matched are shown, and each count is the results in front of you, so the row tells you the shape of the answer before you’ve read one. Four decisions and one note is a different situation from one decision and fourteen notes.
Twenty-five results is the most it puts on screen at once, and when there are more it says so, in the results, with the real total: Showing the 25 closest matches of 61. Add another word to narrow it down. When nothing matches at all it doesn’t shrug either: it says Try fewer words, or turn on History to include past values, reversed decisions and resolved questions, and hands you the switch inside the sentence.
Open the box with nothing typed in it and it doesn’t ask you to think of something. It shows you two lists: what changed since you last looked, and what you opened most recently.
- ⌘K
- open it, from anywhere
- ↑↓
- move
- ↵
- open the project at that item: the row you picked flashes once, so you can see where you landed
- ⌘↵
- actions for this result: open it in its project, copy it, copy it for an assistant, see where it came from, pin the project
- esc
- close
Past values, if you ask for them. A value you replaced is still findable: turn History on and it comes back, always labeled as past and never ranked as if it were true now. It’s off unless you turn it on, so an ordinary search never returns anything historical, and nothing historical ever arrives unmarked.
What it doesn’t keep
Giraffe is not a transcript archive. It doesn’t store your conversations, and it doesn’t try to remember everything you said. It keeps the conclusions you decide are worth keeping, and nothing else.
Nothing is saved on its own. Giraffe never reads your chats in the background; it only ever sees what you explicitly ask an AI to send it. If you don’t save something, it was never there. You can read, edit or delete any project yourself, any time, with no AI involved.
There’s no link back to the original conversation, either. Assistants rarely know the address of the chat they’re in, so a source is a dated record of the update, not a door back to it.
What it doesn’t search
Your old conversations.
Giraffe has no view of your chat history and no way to read it, not in ChatGPT, not in Claude, not anywhere. There’s no import yet either, so search covers what you’ve saved and nothing from before that. If you came looking for a way to index three years of ChatGPT, we can’t do it, and you should know that now rather than after you’ve made an account and found an empty box.
Anything you didn’t save.
Nothing is captured automatically. What’s searchable is what you decided to keep, which is the same reason the results are short enough to read at a glance instead of being passages you have to skim.
What a connected assistant can’t do
Connecting an assistant doesn’t hand it your account. It can create projects, list them, read one, search across them, read a project’s history, and update what’s in it. What it cannot do is the part worth knowing: it cannot delete a project, rename or archive one, create or rename the spaces you file projects into, see or change your connections, or pin a project to the top of your library, not as a policy we promise to honor, but because those permissions were never granted to it in the first place.
End a connection and it stops immediately, not at the next refresh, not after a delay. It stays refused until you allow it again. Every connection you have is listed in Settings, under Connections, each with its own control to end it.
Access is enforced by the database itself: your account can only read or write its own rows, whether the request comes from you or from an assistant you’ve connected. One thing we won’t dress up, though: this isn’t end-to-end encryption. Giraffe staff with database access could technically read stored content. We’d rather say that plainly than let you assume otherwise. The longer version is on our privacy page.
Start with one project. It stays yours: read it, edit it, or delete it whenever you like.