Chat with Nova

Nova is your main assistant. You talk to it in the chat on the main screen. From the same composer you can also talk to any agent or team, choose the model, and decide how much Nova may do on its own.

The composer

The controls under the message box, from left to right:

Press Enter to send and Shift + Enter for a new line. The placeholder tells you who will answer: Message Nova…, or the name of the agent or team you picked.

Thinking effort

Click the effort button to open a slider with six steps: Low, Medium, High, Extra High, Max and Ultra. The default is Medium. Higher settings take longer and reason more; from High up, Nova can also take extra private thinking passes before it acts. The setting is saved for your account.

RAG and the Memory Explorer

With RAG on, Nova leans on your knowledge, documents and memory for that chat. Its memory of past conversations stays available either way. While RAG is on, hover the database icon to open the Memory Explorer and tick which memory folders may be searched in this chat. Select all, Deselect all and Search folders… help you choose. The scope is saved for that chat.

Work Mode and Multi Agent Mode

• Work Mode is for longer jobs. Nova breaks the work down and may ask a few clarifying questions first. It proposes a plan for your approval, hands parts to helper agents, and checks the result before it answers.
• Multi Agent Mode turns the chat into a team workspace for the selected team. If you have no team yet, it says Create a team first. See Teams.
• Deep Reasoning switches deep reasoning on or off for the chat. Steps it produces are labeled Deep Reasoning in the work log.

Attachments

Click the paperclip, drag files anywhere onto the chat (Drop files to attach them to your message), or paste an image. Previews appear above the box; × removes one, and clicking an image opens it larger. The same file is never attached twice.

Writing a long message

Hover the right end of the box and click expand to open Compose a message, a large editor with Undo, Redo, Clear and Wrap text. Back to chat (or Ctrl + Enter, or Esc) returns with your draft; it does not send.

Choose who answers

The picker has four tabs and a search box.

Slash commands

Type / at the start of the box to open the command palette. Move with the arrow keys, pick with Tab or Enter, and close it with Esc. Replies appear in the chat in a SyntrofAI Command box. The palette shows a description for each command.

Slash commands cannot be sent together with attachments.

While Nova works

When you send, your message appears at once, followed by Sending message and Starting agent. The status line under the chat then shows the live step, and the answer bubble shows one line on what Nova is doing at that step. When the answer arrives, it replaces that line.

You can keep typing while a reply runs. Pressing send queues your message under Queued Messages (N) above the composer, and queued messages go out one by one when the reply ends.

• Send now on a queued message hands it to Nova while it is still working, as an update to the running job.
• Send all queued sends them together as one message. Clear all or Remove drops them.
• The queue is saved with the chat.

To stop, press the red square with the box empty. With text in the box, the same button queues the text instead. After you interrupt a reply, pressing send with an empty box resumes it. If a message cannot be sent, your text and attachments come back into the box.

When background tasks are running, a small panel above the composer shows them. Expand it to see each one's age, or end one with × (Kill task).

Reading an answer

Each reply has a work log above the answer: numbered steps such as Agent, Tool, Code Run, Browser, Utility, Memory or Deep Reasoning, each with a short preview (the search query, the code, the address) and a tick when done. Click a step to open it. The log shows the latest 20 steps; Load 20 earlier steps shows more.

Some tools draw their own cards:

• Web search shows how many sources it found and a card per source.
• Text editor shows the file and how many lines were added and removed.
• Request map (from Understand) shows how clear the request is, what is known, what is still missing, and choices with a Recommended tag.
• Plan and Browser Automation have their own cards, and browser steps can show a carousel of screenshots.
• Platform Control and Widget Control show a status: Awaiting approval, Pending, Not executed, Failed or Completed.

The answer itself can contain tables, code blocks with Copy code, images you can enlarge, task lists, diagrams, charts, videos and tabbed galleries. Links open in a new tab.

Plans and approvals

For larger jobs Nova proposes a plan before it starts. The plan card sits above the composer with its state (Plan in progress, Plan awaiting approval, Plan paused, Plan complete) and a progress count. Minimize it to a small Plan in progress pill.

The dialog header shows a clock, Approves as proposed in and the time left. If nobody answers before it runs out, the plan starts as proposed.

When Nova wants to change something in the product for you, such as creating a schedule or editing a workflow, it shows a Nova needs your OK card first. Nothing changes until you press Approve: these cards have no clock and wait for you. Cancel leaves everything as it is. Some approvals also arrive as toasts with buttons.

When Nova needs your input

If something important is unclear, Nova stops and asks before it acts. A pop-up headed Nova needs your input opens, titled with how many decisions it needs, such as 2 decisions before Nova continues, with the objective underneath. When there is nothing to ask, it reads Confirm Nova's understanding instead.

• The questions, one card each: Question 1 of 2, the question, and why it matters. Pick an option, or tick several when more than one can be true. Nova's pick carries a Recommended tag with its reason. When an answer of your own is allowed, type it in Your own answer. The bottom corner counts your answers.
• Nova's understanding, on the right: its Confidence, and the Components, Known, Assumptions and Still unresolved points, each with a count. Edit lets you correct the objective, the components, the lists and the confidence; Done finishes.
• The buttons: Submit answers sends your answers once every question is answered (Confirm understanding when there are none). Use recommended answers every open question with Nova's pick, keeps the ones you answered, and sends. Skip carries on without answers, and Nova treats those points as uncertain.

The clock at the top, Recommended answers apply in 1:59, counts down from two minutes. When it reaches zero, Nova's recommended answers are applied, the pop-up closes, and a notice tells you so. − or Esc minimizes the pop-up into a Nova needs your input chip beside the plan and loop chips, with your answer count and the clock still running. Click the chip to pick up where you left off. On WhatsApp the same questions arrive as a message; with no reply within 2 minutes, Nova's recommended choices apply.

Loops

A chat can run loops: work that repeats on a cadence. A Loop or Loops (n) chip shows their state. Click it to see this chat's loops with their cadence, timezone, next and last run, and the buttons Edit, Run now, Pause or Resume, and Stop.

The Command Center

Open it with the sliders icon in the composer.

• System prompt opens an empty box for your own standing instructions, added after the built-in ones. Save it for Session Only or Global.
• Import knowledge takes .txt, .pdf, .csv, .html, .json, .jsonl and .md files and adds them to memory.

Long chats

Very long chats are compacted so they stay fast. The model writes a hand-off of the older part of the conversation and keeps the recent turns word for word. Every message stays visible to you.

• Compaction runs on its own at a threshold you can set in Settings → Agent → Chat Model (Compact the chat at).
• To run it now, use Compress chat in the Command Center. The Compact Chat dialog shows the current size, the threshold, how much is kept, which model writes the hand-off, and the last result. Press Compact now. It cannot run while Nova is working.

Hover Context Window at the right end of the status line to see how full the chat is. It shows the model's window, what share goes to instructions, history and extra context, and how far the chat is from automatic compaction.

Message controls

Under each finished answer: the model's name and logo, ⋯ for details such as the date, Copy text, Speak text, and for answers of 100 words or more, the word count and how long it took.

Under your own messages: Copy, Speak, and Rewind conversation to this message. Rewinding deletes everything after that message and puts the message back in the box. It asks first, and it cannot be undone.

Voice

The microphone starts a live voice conversation with Nova. Its color shows the state: amber while connecting, green while listening, red while recording, purple while processing. A glowing sphere appears at the bottom while the microphone is on. While Nova speaks, a red square lets you interrupt and talk.

• The first time, your browser asks for microphone permission.
• Language, personality and a custom style live in the picker's Voice tab.
• Engines and voices live in Settings → Agent → Voice. With the classic engine, the microphone dictates text into the box instead.

Chat links

Each chat has its own link, so you can bookmark it and reopen it later. It works for people signed in to the same account. The browser's Back and Forward move between the chats you opened. To keep a copy outside the app, use Export in the sidebar for a printable transcript.

Related