Is it worth saving something I might only need once?

You can’t tell in advance which one that is, so the useful test is smaller: would it annoy you to find this twice? An annual pruning window, a deduction you use every January, the settings that finally made something work — those come back. Saving one is a single sentence to ChatGPT or Claude, and nothing is saved unless you ask for it. Giraffe has no view of your conversations.

How is this different from searching my ChatGPT history?

A chat history is a list of conversations ordered by when you had them, and it only covers the assistant you had them in. A Giraffe project is organised by what’s true: one current value per fact, the decisions you settled, the questions still open. You read where things stand instead of reconstructing it from thirty conversations, and one word searches across every project you keep.

What happens when something I saved turns out to be wrong?

You tell your assistant the new value, or type it into the project yourself, and that becomes the live one for that row. The old value isn’t overwritten — it stays behind the current one with the date it was replaced and whether you or an assistant wrote it. Giraffe never notices this on its own and never asks you to confirm a value; a fact changes when you say it changed.

Can I read what I saved without opening ChatGPT or Claude?

Yes. A Giraffe project is a page in your browser — the current facts, what you decided, what’s still open, in the words you saved them in. Most of the time reading it is the whole job and no AI is involved at all. The project is yours alone; nobody else can open it, and you can delete any of it whenever you like.

Everyday learning

Keep the things you look up.

You look things up with ChatGPT or Claude all day — a tax rule, a plant that keeps dying, how a visa actually works — and keep almost none of it. None of it feels like a project at the time. It’s a two-minute answer you’ll want again in six months, by which point you won’t remember which conversation it was in, or which assistant you asked.

You keep working in ChatGPT or Claude the way you do now. Giraffe is where the answer is still there the second time you need it.

Why you keep looking up the same thing

Finding out stopped being the hard part. You ask, you read two paragraphs, you understand it, you go and do the thing. The tab closes behind you and none of that feels like a loss, because the answer took ninety seconds and you can always get it again.

Eleven months later the same question comes back — the same shrub, the same form, the same deduction — and the conversation that answered it is gone. You know you asked. You might remember it was spring, or that it was Claude and not ChatGPT. Neither of those is a search term. So you ask again, read the same two paragraphs, and act on an answer you already owned. Do that with the hundred-odd small things a year takes and you have spent a working day re-finding what you already found.

Where an answer goes after the chat ends

When you land on something you’ll want again, you tell the assistant to write it down: “Add that to my garden project — the hydrangea flowers on old wood.” That sentence is the whole of it. Nothing else from the conversation follows it in, because nothing is captured on its own.

It lands in a project you can open in a browser and read on its own — what’s true now, what you settled, what’s still open — in the words you saved it in. These projects are small and you keep several: one for the garden, one for the house, one for the paperwork of moving countries. You don’t have to remember which one an answer went into, because one word finds it across every project you keep. Most of the time that’s the whole job, with no AI session at all. Here is how Giraffe works, step by step.

Three years of a garden, on one page

Say the garden came with the house and you’ve been working out what’s in it ever since. A question in March about a shrub that didn’t flower. A photo identified in July. An evening in January on rootstocks, because the apple turned out to need a pollination partner nearby. Thirty-odd small questions across three years, not one of them big enough to open a notes app for. Saved as you go, they come to this:

The garden

What’s planted here, and when to touch it

Updated in March

Current

Hydrangea: The mophead by the shed flowers on old wood — dead-head in spring, never cut it back
Wisteria: Twice a year: long whips back to five leaves in late July, back to two buds in January
Apple: ‘Discovery’ on M26 — pollination group 3, and the crab two doors down does the work
Hedge: Beech, not hornbeam — it holds the brown leaves all winter, so it gets cut in August
Soil: pH 7.2, chalky — blueberries go in a pot of ericaceous compost, never in the border

Decisions

The fig on the north wall goes. Three summers, no fruit, and that wall gets about four hours of sun in June.
No feed on the lawn. The clover is doing something for the bees and the grass has been fine without it for two years.

Open questions

Whether the apple is really a ‘Discovery’. The label came with the house, and it ripens a good fortnight later than it should.

Notes

Every pruning calendar I read is written for a warmer garden than this one. The bottom of the slope is a frost pocket and runs about three weeks behind the top.

Recent changes

In March · You

HydrangeaCut back hard in early springFlowers on old wood — dead-head only

Last July · Claude

SoilpH 6.5, from a strip testpH 7.2, from a proper sample

Nothing is captured automatically. Giraffe has no view of your conversations — only what you ask your assistant to save, or type in yourself.

Almost nothing here changes. That’s the point.

Look at the date on the card. Updated in March, five months ago. On a project holding a renewal or a move, that stamp would mean it had gone quietly out of date and you’d stop trusting it. Here it means the opposite: those five rows were right in March, they are right now, and this is the page you open next January when the wisteria wants its second cut. A fact about how a plant flowers doesn’t need maintaining. It needs finding.

Two rows have moved in three years, and both of them are the same kind of move: you were wrong, not the garden. You cut that hydrangea back hard one spring on advice written for a different sort of hydrangea, and got a summer of leaves and no flowers. The pH came off a strip test before it came off a proper sample. When you correct either one, the new value becomes the live one for that row and the old one stays underneath it, dated, with whether it was you or an assistant who wrote it down. Giraffe didn’t catch the mistake — it doesn’t watch your garden and it doesn’t watch your chats. You said so, and a fact is only allowed one current answer.

A tax rule, a chord shape, a visa

The garden is one shape of this. The others are identical from the inside. The deduction you worked out last January and will want again next January. The tuning and the two chord shapes that took an hour to get under your fingers. The order a visa wants its forms in, and the one that has to be certified before it’s sent. Each of those is an hour the first time and four minutes the second — if you can find it. A few will grow into something with dates and a budget, and belong beside a plan with a lot of moving parts. Most of them just want somewhere to sit.

You can’t tell at the time which of them you’ll need again, which is why “is this important enough to save” is the wrong question. The one that works is smaller: would it annoy you to find this twice?

A few more questions

Is it worth saving something I might only need once?: You can’t tell in advance which one that is, so the useful test is smaller: would it annoy you to find this twice? An annual pruning window, a deduction you use every January, the settings that finally made something work — those come back. Saving one is a single sentence to ChatGPT or Claude, and nothing is saved unless you ask for it. Giraffe has no view of your conversations.
How is this different from searching my ChatGPT history?: A chat history is a list of conversations ordered by when you had them, and it only covers the assistant you had them in. A Giraffe project is organised by what’s true: one current value per fact, the decisions you settled, the questions still open. You read where things stand instead of reconstructing it from thirty conversations, and one word searches across every project you keep.
What happens when something I saved turns out to be wrong?: You tell your assistant the new value, or type it into the project yourself, and that becomes the live one for that row. The old value isn’t overwritten — it stays behind the current one with the date it was replaced and whether you or an assistant wrote it. Giraffe never notices this on its own and never asks you to confirm a value; a fact changes when you say it changed.
Can I read what I saved without opening ChatGPT or Claude?: Yes. A Giraffe project is a page in your browser — the current facts, what you decided, what’s still open, in the words you saved them in. Most of the time reading it is the whole job and no AI is involved at all. The project is yours alone; nobody else can open it, and you can delete any of it whenever you like.

Look it up once.

Start with Giraffe

Free. An email code gets you in — no password to invent, no card, nothing to install.

Works on every Claude plan — Free allows one custom connector at a time, and on Team or Enterprise an owner has to add it first. ChatGPT needs a paid plan, on the web, with Developer mode turned on, and it won’t work inside a ChatGPT Project. Reading works there; whether saving does depends on your plan and your workspace, and ChatGPT asks you to approve each save by default. Gemini works through Spark: add Giraffe as a custom app on the Gemini web app (a personal Google Account, set up on the web; availability depends on your Gemini plan and Google’s rollout), then use it from Spark rather than an ordinary Gemini chat. See the setup guides.