Changelog

Version 2.0.9 (April 24, 2025)

  • New Logo!
  • Chats no longer auto scroll to the bottom. Instead there is a button to scroll to the bottom.

Version 2.0.8 (April 23, 2025)

  • Experimental Image Generation tool. Adds support for gpt-image-1 in multi-agent mode.
  • Try things like "Generate an image of a comic about the blog I'm reading".
  • I may or may not remove this feature, would love to hear if it's useful or not!

Version 2.0.7 (April 22, 2025)

  • Add a new prompting guide to help you get started with Do Browser.
  • Improved reasoning and tool use for multi-agent mode.

Version 2.0.6 (April 22, 2025)

  • Make smart mode the default model.

Version 2.0.5 (April 18, 2025)

Classic Mode

  • o4-mini has replaced o3-mini as the default model.

Multi-Agent Mode

  • Add a fast and smart mode to the model selector. This changes the reasoning effort for the model.

Version 2.0.4 (April 15, 2025)

  • Add more information about Google Sheets parsing so the agent can understand when you refer to a specific cell.

Version 2.0.3 (April 15, 2025)

  • Reasoning improvement for multi-agent mode.
  • You can now copy messages from the chat.

Version 2.0.2 (April 15, 2025)

  • Fix a bug where Do Browser would not let you accept multiple permissions in a single session.
  • Do Browser now properly handles the openTab command.

Version 2.0.1 (April 14, 2025)

Classic Mode

  • GPT 4.1 and GPT 4.1-mini are now supported.
  • GPT 4o, GPT 4o-mini, and Claude 3.7 Sonnet have been removed from the list of supported models.

Multi-Agent Mode

  • New "think" tool to help with planning.

Version 2.0.0 (April 14, 2025)

Introducing the new Doobie Multi-Agent mode. The pre-existing agent is now called Classic mode. Multi-Agent uses a mixture of tools, agents, and models to help you accomplish your tasks. These capabilities include:

  • Searching the web
  • Searching your browser history
  • Listing and opening tabs
  • Recording memories
  • Controlling the tabs

Multi-Agent mode tends to be much better at general purpose inquiries and will help you plan out the steps that you need to take rather than rely on you crafting the perfect prompt. In particular it performs much better at data extraction than Classic mode.

Examples of things that work much better in Multi-Agent mode that are nearly impossible for Classic mode:

  • Open the 8 videos in this Youtube playlist and craft a newsletter article about them.
  • Analyze the content of a complex webpage and extract specific information.
  • Draft a LinkedIn post in my style and tone about this article and post it to LinkedIn.

In particular one of the most exciting features is the new memory capabilities you can do things like:

  • Go to your twitter page and ask doobie to "Extract the style of my tweets and example ones and remember how my tweeting style for the future"
  • "Look at my top websites and personalize yourself to me based on that"

These features have made Doobie my main replacement for ChatGPT / Claude now!

On the flip side the automation capabilities of Multi-Agent mode are a bit different than Classic mode. It uses a combination of the old Do Browser agent and Open AI operator to interact with websites. Overall it will be a bit slower, but more accurate. Your old prompts will probably not work as-is with Multi-Agent mode, but you can switch to Classic mode if you need to.

Full Release Notes:

  • You can switch to multi-agent mode by clicking the new mode selector in the top left of the app.
  • Multi-Agent mode is better at general purpose inquiries. It will plan out steps to take before immediately executing them.
  • Multi-Agent mode uses a mixture of tools, agents, and models to help you accomplish your tasks. Including Gemini 2.5 pro, Open AI Operator, and Gemini 2.0 flash. These are the current tools and agents in the release:
    • Read a tab (Use the content extractor agent)
    • Control a tab (Use the browser control agent)
    • List Tabs
    • Open Tab
    • Search Web (Use the web search agent)
    • Search History
    • Record Memory
    • Get Top Sites
  • You can manage your memory by clicking the brain icon in the upper right corner of the app.
  • Multi-Agent mode now supports file attachments, you can now attach files like images, pdfs, and more for context.

Version 1.3.44 (February 28, 2025)

  • Add a way for people to add logs to bug reports.

Version 1.3.43 (February 27, 2025)

  • Added support for GPT-4.5-preview. This is only available if you bring your own key.
  • Do Browser can now switch tabs, to perform tasks.
  • Google sheets insert is more reliable (it won't get stuck on special characters or newlines).
  • Do Browser now has a free trial.

Version 1.3.42 (February 24, 2025)

  • Added support for Claude 3.7 Sonnet (recommended).
  • Reorganized the model selector in settings.

Version 1.3.41 (February 20, 2025)

  • Allow the agent to do an "analyze" step which should help it no longer say that it is going to analyze something and then just pause.
  • The agent should be a bit better at actually counting the steps if you give it a number of things to do.
  • Errors where gemini wouldn't be able to parse output should be fixed.
  • You can now save a run as a saved prompt for later. If you have a run that works well you can save it and it will capture information about the run so it should be more deterministic next time.
  • New feature: Context. You can now add past saved prompts to context so can build on past runs that worked well.

Version 1.3.40 (February 17, 2025)

  • Gemini 2.0 Flash is now the default model.
  • There is a new modal for when you are offline.
  • There is now a button to clear all history.
  • Fix a bug where you would get a 400 if you switch between a model that supports images and one that doesn't.

Version 1.3.39 (February 6, 2025)

  • Gemini 2.0 Flash is now available as a model.
  • Tables correctly render in the chat.
  • Model should be less likely to say they are going to analyze something and not do it.

Version 1.3.38 (February 1, 2025)

  • o3-mini is now available as a model.

Version 1.3.37 (January 31, 2025)

  • Doobie now hovers elements before clicking them. This fixes issues on sites like Instagram.

Version 1.3.36 (January 29, 2025)

  • Asks for feedback at the end of a series of steps.
  • Increase the number of steps you can do in a day.

Version 1.3.35 (January 28, 2025)

  • Now picks up more options in combo boxes.
  • Library updates.

Version 1.3.34 (January 27, 2025)

  • Fix issue where the api would crash if your monitor was too high resolution.

Version 1.3.33 (January 27, 2025)

  • Boxes no longer flash on the screen with screenshots enabled.
  • New onboarding cards with sample prompts.
  • Ability to bring your own DeepSeek key.
  • Saved prompts. You can now save prompts and use them for later.

Version 1.3.32 (January 24, 2025)

  • Fix issue where the button that said doobie was controlling a different tab would stick around.
  • Add a toggle to settings that lets you toggle off sending personal info to the model.
  • Fix issue where screen would flash and selectors would close.

Version 1.3.31 (January 22, 2025)

  • DeepSeek R1 is now available.

Version 1.3.30 (January 21, 2025)

  • Screenshots now enabled by default! This makes Doobie much smarter.
  • Added links to the ToS and Privacy Policy in the Settings.
  • You will now automatically log into the extension if you are logged into the website.

Version 1.3.29 (January 19, 2025)

This is an update with a lot of behind the scenes changes, prepping for the 1.4 release. It includes a rewrite of the main agent to make it more stable and reduce common bugs.

  • Includes updated rate limits again (now 100 premium steps per day).
  • Add a link in the Settings to change the keyboard shortcut to open the Side Panel.
  • Your past commands you've entered will now show up in the address bar when you type "do".
  • The animated background has been reenabled.

Version 1.3.28 (January 16, 2025)

  • Reduce hallucination where Doobie would try to write to a google sheet or doc when it wasn't on the right url.

Version 1.3.27 (January 15, 2025)

  • Change to prompt to improve rate limits and load times.

Version 1.3.26 (January 9, 2025)

  • Increased usage limits for premium models. You now get 50 steps per day. I'm monitoring usage to see if I can afford this.

Version 1.3.25 (January 8, 2025)

This update makes some changes to how Doobie controls tabs. Doobie used to lock to a tab once you send a command while that tab was open. This was cool in that you could talk to a tab that was in the background, but the behavior was pretty confusing to many people.

This updates overhauls the tab management system to unlock some things like letting Doobie choose what tabs to control in the future.

  • Doobie now always operates on the active tab.
  • If you switch tabs while Doobie is trying to execute a command it will pause and wait for you to switch back.
  • There is a clear indicator in the UI on what tab Doobie is operating on.
  • Doobie will now follow tabs if it clicks a link that opens a new tab.

Version 1.3.24 (January 7, 2025)

  • Add a refresh button to the paywall modal so you can force it to check subscription status.
  • Make cheaper models more robust to formatting errors.
  • Remove avatars in the chat for more screen real estate.

Version 1.3.23 (January 6, 2025)

  • New feature quick replies: Doobie will now give you a list of follow up actions you can click rather than having to type things.
  • Fixed a regression where Doobie would get stuck in a loop sometimes.
  • Fixed a bug where you wouldn't be able to load the extension in Arc.
  • Add a new button to disable all extensions if we still hit an extension conflict and aren't detecting the conflict.

Version 1.3.22 (January 3, 2025)

  • Fixes error where model would sometimes output incorrect JSON format.
  • Adds very basic support for the Arc browser. Note that typing "do" in the url doesn't work and chat shows up as a window rather than a side panel.

Version 1.3.21 (December 31, 2024)

  • Deepseek Chat is now available as a basic model.

Version 1.3.20 (December 30, 2024)

  • Add install screen
  • Fix bug where invoking the extension by typing "do" in the url bar would not work

Version 1.3.17 - 1.3.19 (December 30, 2024)

  • Add version history to the extension
  • Your chats are now persisted when you open and close the extension and you can go back to past chats.

Version 1.3.16 (December 25, 2024)

  • Some copy changes to the extension.
  • Added a link to the extension page in the app.

Version 1.3.15 (December 23, 2024)

  • These update notes are now in the extension 🙂
  • Youtube captions are now disabled by default. You can enable this feature in settings.
  • New Included Basic Model: Claude Haiku 3.5. Please try it out and let me know what you think!
  • Message History (lite). You can now quickly go back and forth between last sent messages by pressing the up and down arrow keys.
  • Removed Gemini 2.0 Flash from the open router models since the rate limits are too low.

Version 1.3.14 (December 21, 2024)

  • The TextArea now grows automatically as you type.
  • The stop button is now red.

Version 1.3.13 (December 21, 2024)

  • Fixed a bug where you would receive an error after stopping a generation and talking to Doobie again.
  • Each step should cost less tokens now. You will feel this if you are bringing your own key. Part of the rollout to update the number of steps.

Version 1.3.12 (December 19, 2024)

  • Updated planning is re-enabled. llama 3.3 and gemini 2.0 flash are supported from Open Router.
  • Fixed a perf bug where the extension would slow down on some large websites.

Version 1.3.11 (December 19, 2024)

  • This reverts the new planning overhaul temporarily while I fix some kinks with it.

Version 1.3.10 (December 19, 2024)

  • Better detection of conflicting extensions (it will pick up lastpass and a few others)
  • Fix issue where large google sheets crash the extension

Version 1.3.9 (December 18, 2024)

  • This update is a huge planning overhaul.
  • Do Browser should be a bit better at not getting lost at mid-length (10-15 step) tasks.
  • The UI has been updated to show what Doobie is actually doing.
  • More importantly this update makes cheaper models a lot more usable. GPT-4o-mini is quite usable now.

Version 1.3.8 (December 15, 2024)

  • If you have a chrome extension that is preventing Do Browser from reading the page you now will get a modal that will let you know which extension is conflicting and let you disable it.

Version 1.3.7 (December 14, 2024)

  • Basic support for reading and writing Google Sheets - Note it only works on smallish spreadsheets at the moment.
  • Basic support for reading and writing Google Docs
  • Better short term - memory. Ask it to "memorize" something on a page so it won't truncate it as the context gets too long.

Version 1.3.6 (December 13, 2024)

  • Attempted Fix for the Protocol Error issues
  • Attempted Fix for the Session ID not found error
  • Made the discord easier to find in the app
  • Made it clearer what email you are logged into (you can check in settings)

Version 1.3.5 (December 12, 2024)

  • Gives me the ability to push a notice from the server (in case open ai is down lol)
  • Fixes a bug with edge where it would say "Cannot access edge:// url" when invoking from the new tab screen
  • More crash telemetry

Version 1.3.4 (December 11, 2024)

  • The "debugger is already attached to tab id x" bug should be fixed now
  • The random 500 error that would sometimes occur should also be fixed
  • The manage subscription button should take you directly to the page without sending you an email that you have to open first

Version 1.3.2 (December 9, 2024)

  • You can now bring your own Open Router Key. I recommend this over bringing an OpenAI key if you need more credits. Right now I only support gpt-4o, gpt-4o-mini, sonnet-3.5, and haiku-3.5. Haiku 3.5 is probably the best bang for buck right now! I was really impressed. If there are more models that you want to try let me know, the main restriction is it needs to support tool calls
  • There is a new button to change what tab Doobie is controlling mid conversation! There is still work to do on making the ux of this better, but this is my quick way to unblock you all for now.
  • I fixed a perf issue where db would cause a cpu spike while it was open. It should be much snappier now
  • I rolled out some back of house changes to do Sentry exception capture so I can find these pesky debugger detached errors.

Version 1.3.1 (December 6, 2024)

  • Make error messages clearer (has a reset button now)
  • Make Settings Scrollable

Version 1.3.0 (December 6, 2024)

  • Early Access Release.