Chats and projects

#Projects

Everything you start becomes a project: a folder on your computer with its files, its chats, its memory and its git history. The start page lists your projects in the sidebar. Search it to find a project or an earlier chat.

The sidebar marks each project: a turning ring while Synonode works on it, an amber dot while it waits for your answer, and a blue dot when it finished while you were in another project, until you open it. Start a new project with a tap on Synonode at its top, or Ctrl Shift O. Fold the sidebar with its button and it goes away completely; the button stays at the top left of the chat, and hovering it shows the whole list over the page.

From a project's ⋮ menu in the sidebar you can pin, rename, download it as a .zip, archive or delete it. A project nobody named is named from its work after the first reply. You can turn that off with Name projects in Settings → Capabilities.

#Chats

A project can have many chats. /clear starts a fresh one; the old chat is kept. /resume opens the list of this project's chats, so you can go back to one. /export saves the whole conversation as a Markdown file.

/fork branches the conversation where it is, to try another way without losing this one: you go on in the new branch, and the chat as it was is kept under /resume, marked Forked (give it a name: /fork before the redesign). Only the conversation branches: the project's files are shared, so use restore points or git to go back on code.

Synonode can also look into a project's earlier chats when they're relevant (that project's only, not your other projects'), including by date: what did we decide about the menu on Monday? Turn this off with Search and reference chats in Memory settings.

#Incognito

On the start page, the incognito button starts a conversation that isn't saved, isn't remembered and doesn't add to memory. It's removed as soon as you leave it. Choose Keep this conversation from its menu if you change your mind.

#The message box

• Enter sends. Shift Enter starts a new line.
• Type / for commands and your skills.
• Type @ to pick a file from the project. Its contents go with your message, so Synonode starts with the file in hand.
• Paste or drop pictures and files into the box. On a phone, the + menu also offers your photos and the camera.
• ↑ on the first line brings back what you sent before in this project; Ctrl R searches it.
• Start with ! to run a command yourself: !npm test runs in the project, the chat shows the command and what it printed, and Synonode reads both with your next message, without spending a step running it again.

#The + menu

Item	What it does
Photos and files	Attach pictures or documents. Synonode reads PDFs, Word, Excel and PowerPoint files, and can search inside them. A very long paste (over 4,000 characters or 100 lines) is attached as a text file, so the box stays readable. Text you paste (a log, an email, a page) is marked as pasted for the model: it reads it as material you gave it, and doesn't follow instructions inside it unless you say so. Pictures are numbered in the order you pasted them, so you can say make the first like the second, and each is saved in the project's .synonode/images/ so Synonode can use it, for example as the logo. Send a video and Synonode watches it: it looks at frames across the whole clip, each marked with its time, and writes down what is said in it on your computer, with the Whisper that writes down your dictation. The first video fetches Whisper if the microphone has not yet (about 160 MB, once). It needs ffmpeg; if it is missing, Synonode says how to add it. Ask about one part (what does he say at 1:20?) and it looks at just that span.
Permission	What Synonode may do before asking you. See Permissions.
Research	Turns your next message into a deep research task. See Research.
Thinking	How hard the model thinks before answering. Shown only when your model supports it.

#Suggestions

After a reply, Synonode suggests what you might say next, in grey in the empty box. Press Tab or → to take it, or just type something else. Turn this off with Suggest what to say next in Settings → Capabilities.

#While it works

You don't have to wait. Each step shows as it happens: the files it reads and changes, the commands it runs, the pages it opens. Steps of the same kind are grouped (Ran 6 commands); open a group to see each one.

• Send a message with Enter and it joins the work in progress: a correction, a detail you forgot, the next idea.
• Ask on the side with /btw: /btw which file has the prices? The answer shows above the message box, from what the conversation already holds, without tools, and neither your question nor the answer joins the conversation. It works while Synonode is working: on the helper model if you run one, otherwise as soon as the current turn ends. Close it with its ×.
• Catch up with /recap: one or two sentences on what the conversation has done and what is left, above the message box, without adding to the conversation. /copy copies the last reply (/copy 2 the one before), and /diff opens the Changes view.
• Queue a message with Ctrl Enter (⌘ Enter on a Mac) and it waits until the current work ends, then goes as its own message. Queued messages show above the box, where you can send one now, edit it or remove it.
• Follow the task list. For work of several steps Synonode keeps a list of them, as Claude Code does: it writes every step at once, marks the one it is on and ticks each off as it finishes. If the list goes untouched for 5 steps while it still has steps left, Synonode reminds the model of it (after 10 steps when there is no list, as a hint to start one). The list is one card at the very end of the chat, under what is running now, and scrolls away with the rest of the chat when you scroll up: an empty circle for a step to do, a turning ring for the one it is on, a check for one done. It goes once every step is done. The working line says the step in its own words. The list survives a long conversation being compacted, and /todos prints it.
• Stop with the stop button or Esc: it stops at once, even in the middle of the model's thinking or a long command. What was written so far stays in the chat, and your next message starts as fast as ever.

#Replies

How much of the model's working you see is the transcript view: Normal folds the tool calls into summaries and shows the full replies; Thinking adds the model's thinking, each Thought for… open between the folded tool calls (it is offered once the conversation has thinking); Verbose opens every tool call, file read and step, plus the thinking. Switch it with Ctrl+O or Settings → Code → Default transcript view.

While the model thinks, the working line counts its thinking; on Thinking or Verbose it shows the newest line of it as it is written, and the opened work shows all of it. On a computer, pointing at a reply shows how long ago it was written (Just now, 5 minutes ago); on a tablet it is always there. On a phone, a reply has only Copy and Read aloud, as in the Claude Code app. Beside the message box's buttons is how fast the model writes (27 tok/s): while it writes, and after, the speed of its last call; hold the mouse on it for the reading speed too. Turn it off for one device, here and on the agents in Background tasks, in Settings → General → Appearance. On a computer, point at one of your messages (on a tablet, it is always under it) to see whether Synonode has read it (two checks; one while it waits), when you sent it, and to copy it or Rewind to just before it: the conversation, the code or both, with your words back in the message box to send again or change.

Ask Synonode about itself too: how do I comment on a line?, can you work on a schedule?. It reads these docs to answer, so it tells you where things really are.

When you write again after a pause of half an hour or more, a wavy line with the time sits above your message, so you can see where you picked the conversation back up.

Scrolled up to read earlier messages? A round ↓ above the message box takes you back to the latest.

#Charts, diagrams and files in the chat

Synonode shows things right in the reply, so you see them with the answer, on a phone too:

• Charts and small interactive visuals, all in one clean style that follows your theme: a chart of the numbers it just gave, a calculator, a slider to try an idea. Expand shows one full size, Download saves it as a page, and a button inside one can ask Synonode a follow-up for you.
• Diagrams of a flow or a structure, which you can also expand or download as a picture.
• Formulas, written as in a textbook.
• Files it made for you, each as a card with a download: pictures, sounds and videos play in it, a page runs in it (a long one shows its start, with Show more for the rest and a button to see it full size), a table shows its first rows, a short document reads as a page (its start, with Show more for the rest), the lines of code it means are highlighted, and a small change shows as it is. It sits in the reply like a chart does, with small buttons under it: open it in the preview, see a picture full size, and download it.

When Synonode looks at a picture (a page it built, a screenshot you sent, a picture in your files), you see it too: large, under the step that looked at it, up to three with a +N for the rest. Tap one to open it full size.

Right-click any picture in the chat (or the one open full size) to Copy image or Save image; on a phone, press and hold it.

To point at something in a picture, tap it to open it full size and choose Draw: mark it up in red, then Add to message puts the marked-up copy in the message box, ready to send with what you want changed. The picture in the chat stays as it was.

While Synonode makes one, a row says so (Chart · building · 40 lines), as it does for its other work, instead of its code going by. Synonode looks at each chart and diagram as your phone would draw it. One that doesn't work is written again before the turn ends, and you only see the version that works.

The preview opens by itself only for what needs the room: your running app, a big page, a long file, a change across many files. A sound Synonode makes is a real audio file that plays in the chat and downloads as itself.

If a turn fails, for example because the model stopped answering, a note stays in the chat saying what happened, with ↻ to try again. Nothing is lost.

#Output style

/output-style, or Output style in Settings → Capabilities, changes how Synonode talks while it works:

• Default: short and to the point.
• Explanatory: adds a short insight about the choices it makes.
• Learning: also leaves small parts for you to write, marked TODO(human).

#Long conversations

A model can only read so much at once (its context). /context shows what fills it. While a reply is being prepared, the line under the chat says how much it's reading and roughly how long that takes.

When a conversation nearly fills the context, Synonode summarizes it and carries on (Auto-compact). You can also run /compact yourself, optionally saying what to keep: /compact keep the API decisions. With Auto-compact off, you're warned instead and choose when to compact. Where it happened, a wavy line across the chat says so and how many tokens it saved: Compacted session · saved 173.4k tokens.

"Nearly fills" leaves room for the summary itself, the Autocompact buffer in /context. Synonode learns that room from the summaries your model really writes: it starts at about 14k tokens and shrinks to the size of your model's recent summaries, with a margin. Compaction never starts later than 90% of the window. A summary that runs out of room makes the buffer bigger again.