Conol 1.6.0 — Conol on Windows
Conol
- release-notes
Until now the desktop app was for the Mac only. This release brings it to Windows. It also fixes a handful of things that got in the way of the agent finishing what you asked it to do.
The desktop app, now for Windows
Why. The desktop app keeps Conol in its own window with its own notifications, instead of one browser tab among dozens. Mac users have had that since the app launched; many of you work on Windows, and you've been asking for the same.
How to use it.
- On your Windows PC, sign in to Conol in the browser, open the menu under your name and choose Download desktop app. You can also use the download button on conol.ai. Either one detects that you're on Windows and gives you the Windows installer.
- Run the downloaded
Conol-Setup-….exe. It installs just for your Windows account, so you don't need administrator rights, and it adds Conol to the Start menu. - Sign in. Conol opens your browser to sign in and then brings you back to the app.
From then on, Conol updates itself in the background, as it does on the Mac.
Tip
The first time you run the installer, Windows may show "Windows protected your PC". That appears because the installer isn't signed with a Windows certificate yet. Click More info, then Run anyway. You only need to do this once; updates install without the warning.
The app needs 64-bit Windows 10 or 11. Everything you do in it is the same Conol you use on the web and the Mac: your notes, chats and computers are all there.
Long chats stay on track with DeepSeek and Kimi
Why. Every model has a limit on how much of a conversation it can take in at once. When a chat gets close to that limit, Conol summarizes the earlier part and carries on with the summary plus the recent messages. DeepSeek V4.1 Flash and Kimi K3.1 Pro advertise a very large limit, about a million tokens. In practice we saw both of them get unreliable far below it: once a chat passed roughly 200,000 tokens, DeepSeek sometimes returned nothing at all, and Kimi once answered with text that had nothing to do with the conversation. A long task with a lot of reading (dozens of emails, a pile of files) reaches that size easily.
How it works. Nothing to set up. For these two models, Conol now summarizes the earlier part of a chat at around 190,000 tokens (DeepSeek V4.1 Flash) and 200,000 tokens (Kimi K3.1 Pro), well before they start to falter. You'll see a "Context compacted" divider in the chat when it happens. Chats that are already past that size are summarized the next time you send a message. Other models are unchanged.
A side benefit: each new step in a very long chat no longer re-reads the whole history, so long chats on these models also cost fewer credits.
Fixes
- The agent no longer stops halfway without a word. Sometimes a model's reply came back completely empty in the middle of a task, and the chat simply ended there, looking finished. Conol now asks the model again, up to three times. If it still can't answer, the chat shows that the run failed, so you know to try again rather than wait.
- A chat started from Home opens in its place. Sending your first message from Home used to split the window and open the chat beside it. Now the chat takes over the Home tab. Press G then H to get Home back.
- The agent asks for a computer before it needs one. In a chat without a computer, the agent would try a command, fail, and only then ask you for a computer. It now asks before its first command, so the approval card is the first thing you see.
← Back to blog