Chats: new, continued & history

A chat is a single conversation with your Plank assistant. You can keep several going at once, set them aside, and come back to any of them later — nothing is lost when you look away.

Your chats

You can keep as many chats going as you like. Start one with the "+" button ("New chat"). Each chat is one conversation, so you can have one working on a report while another handles something unrelated.

Tap the chat title at the top to see all your chats in this workspace — the most recently active one first. Tap any of them to pick up right where you left off. There's nothing to open or close: every chat stays in the list, so an older conversation is always one tap away.

To remove a chat you no longer need, open that list and tap the trash icon on its row (on a computer it appears when you hover). You'll be asked to confirm — deleting a chat is permanent and can't be undone. Any files the chat produced stay in your workspace; only the conversation goes.

When to start a new chat vs. continue one

A simple rule of thumb: start a new chat for a new topic or a new piece of work; continue the existing chat when you're refining or extending the same work.

Why it matters: a focused conversation gets better, more accurate results than one giant thread that mixes many unrelated tasks. Keeping each chat about one thing also makes your history far easier to navigate later.

One thing to keep in mind: the assistant works on one task at a time, so if you start something new while it's still busy, you may be asked to wait a moment until the current task wraps up. See Workspaces for more on how it handles one job at a time.

Knowing your message was sent

When you send a message it appears straight away, slightly faded, and turns solid once it's safely stored — usually instantly. If the send is taking longer than normal, the faded message breathes gently so you can see it's still going.

If it couldn't be sent at all, you'll see "Not delivered — tap to retry" under it. Tap it to send again.

And if your connection dropped (or you closed the tab) before a message went out, you don't lose it: next time you open that chat, a small note appears above the message box — "1 message wasn't sent" — with Resend and Dismiss. Nothing is sent behind your back; you decide.

Sometimes the assistant's answer is slow to start because the AI service behind it is under load. If that lasts more than a few seconds you'll see "This is taking longer than usual" below the conversation. It almost always sorts itself out — the assistant is still working, and you don't need to do anything. If it drags on, the note will also remind you that you can stop and send again instead of waiting.

Sending while it's still working

You don't have to wait for an answer before writing your next message — send it and it reaches the assistant right away. It's picked up at the assistant's next step, so it won't interrupt a command that's already running: if one is mid-way through, your message waits until that step finishes, however long that takes.

The one exception is waiting. When the assistant is only letting time pass — checking back on a long job, pausing before it tries something again — your message ends that pause immediately, and it picks your message up in seconds rather than at the end of the wait. You'll see it stop on "Waiting" and answer you.

Occasionally there's no running conversation to deliver into yet — while your workspace is still starting up, for instance. Then your message waits as a queued card under the conversation instead — it goes out by itself the moment the workspace catches up, and several queued messages go together as one.

Two ways to take one back:

• The × on the card removes just that message. If it was the only one queued, its text goes back into the message box so you can edit it and send again.
• The stop button stops the assistant and withdraws everything you had queued for that chat. The text comes back into the message box — all of it, in the order you wrote it, with any attachments and anything you'd already half-typed kept at the end. Nothing is sent, and nothing is thrown away.

What the assistant's status means

Just above the message box you'll see a short status while the assistant works. Each state means something specific. The first three are about getting your message there; everything after them is the assistant actually working. On a fast connection you won't see the first three at all — they only appear when a step is genuinely taking a moment:

• "Uploading…" — your attachments are still going up. Nothing has reached the assistant yet.
• "Waking your workspace…" — your workspace was asleep and is starting back up. This is the slow part of a first message after a break.
• "Delivering your message…" — your message is on its way and we're waiting for your workspace to confirm it arrived. If it doesn't, you'll see a "Retrying request" notice — the message is resent for you automatically.
• "Getting to work…" — your workspace has your message and the assistant is picking it up.
• "Thinking" (or "Thinking · 3m" on longer jobs) — it's working, but has nothing to show you yet. The timer is there so a long, quiet task doesn't feel like a stuck one.
• No status at all, with the answer appearing — it's writing. The answer itself is the signal, so the status steps out of the way.
• "Waiting for your answer" — it asked you a question and is holding, not working. It will wait as long as you need. Tap one of the options to answer — a single-choice question sends as soon as you tap — or just type your answer in the message box as you normally would: while a question is open, what you type there goes to the question. Pick "Something else…" if you'd rather write a different answer inside the card.
• "No output for 8m" — it's still connected and still claimed to be working, but nothing has come back for a while. Most of the time it's a genuinely long step and it will finish on its own. If it keeps climbing and you'd rather not wait, press the red stop button next to the message box and send the message again — nothing you've written is lost.
• "Lost contact — recovering…" — the connection to your workspace dropped and it's reconnecting on its own. Your message is safe, and the work usually continues where it left off.

The difference between the last two is worth knowing: "Thinking" counts from when you sent the message, while "No output for…" counts the silence itself. A job that ran fine for ten minutes and then went quiet for eight will say eight — that second number is the one that tells you whether to keep waiting.

A brief network hiccup while nothing is running won't interrupt you — you'll only see a reconnecting notice when there's actually work in progress.

When your workspace is updating

Every so often you'll see "Update available" with a "Refresh now" button above the message box. That's a new version of your workspace waiting to be picked up. Nothing is forced on you — the update only happens when you tap it.

Tapping it rebuilds your workspace, which usually takes under a minute. While it runs you'll see "Updating your workspace…" with a progress bar, a count of seconds elapsed, and what's happening right now: first "Rebuilding your workspace", then "Starting the assistant". It stops as soon as your workspace is genuinely ready — you never have to reload the page to find out that it already is.

You can keep typing the whole time. Messages you send during the update aren't lost and aren't sent into a workspace that can't answer — they wait as queued cards, with a note saying they'll go out when your workspace is ready, and they send themselves the moment it is. You don't have to watch it or retype anything.

If the rebuild itself doesn't finish, the progress stops on its own after three minutes rather than running forever, and "Update available" comes back so you can try again with one tap. And if a message can't be delivered at all, we put the text back into the message box for you rather than discarding it, so you can send it again with one tap.

What "compacting" means

In a long conversation you may see a line saying "Preparing conversation context", with a note that Plank continues automatically. It sits in the transcript like any other step, shows how long it has been running, and turns into "Conversation context ready" when it's done. That's the assistant tidying up the earlier part of the conversation so it can keep going comfortably. You don't need to do anything and you don't need to resend anything — it's automatic, and it keeps long chats fast and reliable rather than letting them get slower and more error-prone the longer they run.

It usually takes a few seconds. If a very long chat needs it more than once, you'll see it more than once — each one is a separate step, shown on its own line.

What it tidies is the bulky raw output of earlier steps — the full contents of a file it read twenty messages ago, for example. The conversation itself, and what it concluded, stay.

And if anything important needs to stick around, it's saved as a file anyway, so it's safe regardless. See Your files.

Choosing how the assistant answers

In the row of buttons under where you type there's a small button showing how the assistant is set to answer — Quick, Standard or Thorough. Tap it to change.

• Quick — fastest. Good for short questions and lookups.
• Standard — a balance of speed and depth.
• Thorough — most capable. Best for difficult or important work.

New chats start on Standard, the balanced middle setting, so you don't have to touch this at all. Whatever you pick stays with that chat, so a chat you use for quick lookups can sit on Quick while another stays on Thorough. Your most recent choice becomes the starting point for the next new chat you open.

Changing this never interrupts anything. The assistant always finishes what it started — the setting only affects how it approaches your next message.

A note on attachments

You can attach files to a chat message (the paperclip, "Attach file") when you want the assistant to look at something just for that one request. For files that should live in your workspace and be available to every chat, upload them instead — see Your files.

On a computer you can also drag files straight onto the chat — anywhere in it, not only onto the message box. The whole conversation lights up with a dashed outline while you're dragging, and letting go attaches the files. Dragging a file from your sidebar onto the chat does something different but just as useful: it adds that file as a reference to your message, so you can say "summarise this" without typing the name.

You can attach up to 20 files per message. Documents can be up to 2 GB each; images are capped at 100 MB, since images are sent to the assistant as vision input, which has its own limits. Each attachment shows its own upload progress, and files upload one at a time — if one fails, only that file retries, so the rest of your message isn't held up.

If you attach an image (a screenshot, photo, or chart), the assistant sees the picture directly and can describe or analyze it — no extra step needed. Just attach it and ask.