OpenBot Mobile 1.4.8
New
- Type
@in an agent's chat or side thread to tag one of the agent's browser tabs or side threads, as on the desktop. The agent gets the tab and can use it, and it can read the tagged thread. A tagged browser tab shows as a badge in the message. Tap it to open that tab. The host computer must have OpenBot 0.62.0 or later. - A tap on a file name in a chat message now opens the file from the agent's workspace or the shared folder. Text, Markdown, code, PDF and spreadsheet files open in the preview, and images open in a sheet. Other files open in the share sheet, where you can save them to Files or open them in another app (#5227).
- Manage the proxies of browser profiles in the mobile app. In Server settings > Browser profiles, an owner or admin can add, change and delete a proxy, such as
socks5://user:password@host:1080, and choose the proxy of each named profile, or Automatic. A saved password shows only as a mask. The list also shows the computers that members share from the desktop app, and shows when a proxy does not answer, refuses the sign-in, or when a computer is offline. Before a change closes open tabs, the app asks. The phone cannot share its own internet connection. - You can change the order of the tools in the server drawer, such as Threads, Schedule and Workflows. Hold a tool and select Edit sidebar, then drag the handles. The order is saved on this device.
- You can unpin a tool to hide it from the server drawer. Hold the tool and select Unpin. In edit mode, the pin button shows or hides each tool.
- Control a computer from the phone. Open the server menu and tap Control computer to see the computer's screen and use it like a trackpad. Pinch to zoom, type with the phone keyboard and the keys above it (Esc, Tab, Ctrl, Option or Alt, Cmd, arrows), and paste text from the phone. The ••• menu has Trackpad mode, Fit to screen, the computer's displays, the quality (Fast at 120 fps, Smooth at 60 fps or Data saver) and connection stats, and the help button shows each gesture. The phone asks for Face ID, Touch ID or the phone unlock first, and stops the stream when you leave the app. The computer must run OpenBot with remote desktop set up.
- On iPhone, the OpenBot share sheet has a message field. Type a message to send with the link, text or file that you share. The message goes before the shared content. If the share sheet cannot send and you open the app, the message is in the chat's message field.
- A server admin can select which models of a provider the model picker shows. Open Server settings › Providers, select a provider, then select Models. The page has a search, Show all, Hide all and Reset to default. Agents can still use every model.
- On the Claude page, a server admin can set how long Claude keeps the prompt cache.
- A server admin can disconnect the server's own sign-in of a provider, such as Claude or Codex.
- A server admin can reset the usage of a Codex account that has usage reset credits.
- The provider pages update when a change comes from the desktop app or another admin, such as the models of the model picker, the prompt cache time or a sign-in. You do not have to open the page again.
- Connect GitLab.com from the Apps marketplace with browser sign-in, so agents can work with projects, issues, merge requests, repositories and CI pipelines. For a self-hosted GitLab, add its MCP server address in the MCP servers settings.
- Connect Omnisend from the Apps marketplace with browser sign-in, so agents can read campaign reports and work with contacts, segments, campaigns and automations. Select read-only access on the Omnisend consent screen when agents only need reports.
- Connect ElevenLabs from the Apps marketplace with browser sign-in to turn text into speech, work with voices, and manage ElevenLabs voice agents. Each generation uses ElevenLabs credits.
- Use OpenBot in Hebrew. The layout mirrors from right to left when Hebrew is the language.
- A workflow output that posts into an OpenBot channel shows as "OpenBot channel". Choose the channel in OpenBot on the computer.
- Settings > Profile now shows your activity on your paired desktop, as the desktop Profile tab does: tokens, peak day, messages, streaks, a year of days, insights, your most active teammates and model usage.
- A menu at the top right of Settings > Profile has Share and Sign out. Share makes the same activity image as the desktop app and opens the share sheet.
- Settings › Appearance has a Message color setting: no color, the agent color, or a custom color that you select from presets or a color picker. A custom color also shows on your messages in channels.
- Settings › Appearance has a Text size slider, from 67% to 200%, as on desktop. It adds to the text size in your phone settings.
- Choose the ElevenLabs model that reads replies aloud, as on desktop: Flash v2.5 (the fastest, and the default), Multilingual v2 or Eleven v3. Open Settings > Voice > Speaking, check ElevenLabs, and select a model in Model.
- A Library in the server drawer shows every picture and file from the chats of a server, newest first, in a bento grid where pictures get the large tiles. You can filter them by type. A picture opens in the image viewer, and you can swipe through the pictures. A file opens in its preview, or in the share sheet. Hold a picture or a file to show it in its chat, share it, or delete it. Pull down to measure again.
- Talk with an agent by GPT live voice, as with Codex live voice on the desktop. Turn on GPT live voice in Settings > Voice and choose its voice. A voice conversation in an agent chat then uses OpenAI's live voice with the Codex (ChatGPT) sign-in of the computer that runs the server. Your voice goes between the phone and OpenAI, and the chat shows your requests and the agent's answers as usual. The computer that runs the server must have a version of OpenBot that supports GPT live voice; other servers keep the voice of the phone. Settings > Voice lists each server as ready, offline, in need of a newer OpenBot, or in need of a Codex sign-in, and an owner or admin can open the server's Codex page from there. When a start fails, the call bar shows the reason that the server gives. It needs this new version of the app from the App Store or Google Play: an update over the air does not add it.
- Use your own OpenAI-compatible voice server, as on the desktop: Speaches, Kokoro-FastAPI or another server with
/v1/audio. Set it up in Settings > Voice > Voice server, then choose it for dictation, for spoken replies, or for both. The phone sends your recordings and the reply text only to that address, so the phone must be able to reach it. - Dictate in English with Parakeet Unified, the English model of the desktop. It runs on the phone, adds punctuation and makes fewer errors than Parakeet v3. Choose it in Settings > Voice > Listening. It needs a phone with 6 GB of memory or more. Whisper turbo and Cohere Transcribe stay on the desktop only: they are too large for a phone.
- The Usage screen of a server shows the CPU, memory and disk use of its machine to owners and admins, in one row at the top. The values update every few seconds while the screen is open. Select the disk to open Storage.
- You can sign in with your company account, such as Microsoft Entra ID, Okta or Google Workspace. Enter your work email: when your company uses company sign-in, its sign-in page opens in place of an email code.
- Dictate with Qwen3-ASR, which runs on the phone. It finds the language you speak, in 30 languages and 22 Chinese dialects, also when you mix Chinese and English in one sentence. Choose it in Settings > Voice > Listening and download the model (about 1 GB). Your speech stays on the phone. It needs a phone with 6 GB of memory or more, and this new version of the app from the App Store or Google Play: an update over the air does not add it. (#5713)
- Read replies aloud with Eleven v4 or Eleven v4 Turbo, the ElevenLabs models that speak more than 90 languages. Open Settings > Voice > Speaking, check ElevenLabs, and select a model in Model. Flash v2.5 stays the default.
Improved
- The ChatGPT code sign-in of a server now tells you to turn on device code sign-in for Codex in the ChatGPT security settings first. For a work account, the workspace admin turns it on.
- When a plan card waits a minute or more before it starts the work, its Start button counts down in minutes, such as "Start · 4:59", not in seconds.
- The Instructions file page of an agent now shows about 14 lines of the file instead of 4.
- On Android 12 and later, the bar at the top of sheets and pages now blurs the content that scrolls under it, and the blur fades out smoothly below the bar. Before, the bar was a solid color. This applies to home, Marketplace, Routines, Usage, Tasks, Workflows, Settings, Create agent and the other sheets.
- The icons for Threads, Schedule, Workflows, Usage and Marketplace in the server drawer and in search are now the same as in the desktop app.
- The Routines page of a server is now called Schedule, as in the desktop app. It shows the calendar of routine runs. You still manage routines in the info of each agent and channel.
- Server settings is now one page. The sections are Server (updates, timezone and memories per agent), People (your role, members, roles and audit log), AI and tools, Data, and Import and transfer. Refresh, Restart and Leave, Delete or Remove come next, and Server key is at the bottom. General and Maintenance are not separate pages any more. A row that is not available for your role does not show.
- The top of Server settings shows the server logo beside its name, with the connection status and the OpenBot version under the name. Owners and admins tap the logo to change it, and tap the name to edit it in place.
- The rows of Server settings and Settings have icons. Where the desktop app has a tab for the same setting, the row uses its icon.
- Profile has the same layout as Server settings: your photo beside your name, with your email under it. Tap the photo to change it, and tap the name to edit it in place.
- Consecutive completed runs of one routine now show as one row in the chat, as in the desktop app. The row shows the routine and the number of runs. Tap the number to show or hide each run. Runs on different days also join, so a daily routine no longer adds a row each day.
- You can now attach a file of any type to a chat on a server that supports it, for example a VPN profile (
.ovpn) or a certificate (.pem). Before, the app refused a file whose extension was not in its list. The agent gets the file and reads it with its own tools. - A voice conversation reads the agent's progress updates aloud while it works, and then the answer.
- A new agent now opens with its own short welcome, in your language, as a question with three first tasks that you can choose or answer in your own words. The setup message that OpenBot sends to the agent no longer shows in the chat. The first task that you write during setup still goes to the agent after its welcome.
- Connect Composio with a sign-in when your server can relay sign-ins. Otherwise, type only your Composio API key. You no longer type an MCP URL.
- The chat list keeps the sections and folders that you close. When you open the app again, they stay closed. Each server keeps its own sections on this device.
- Sign out moved from the bottom of Settings > Profile to the new menu.
- On iPhone, the value and the arrows of a selection row in settings are now gray, as on Android.
- The rows in the info of an agent now have icons, the same as in the desktop agent settings.
- In the info of an agent, Routines shows each routine as a card, as in the desktop app. The card shows the trigger icon, the name, when the routine runs, and a switch at the end. The switch pauses or resumes the routine. Tap the card to open the routine.
- A member of a joined server can see and add its MCP servers, and connect Marketplace apps, when the server allows it. A member changes or removes only the MCP servers that the member added.
- The Roles page of server settings is now named Permissions.
- When you come back to the app after a few minutes, it keeps its connection to your computer and is ready sooner. Before, every return after about 2.5 minutes made a new connection and loaded all servers again.
- When the phone moves from Wi-Fi to mobile data or back, the app renews its connection at once. Before, the connection could stop with no data for up to half a minute.
- A new connection to a server is faster, because the app starts the next steps while it waits for the account service.
- Tables in chat messages look as the desktop data tables do: a rounded card with the header on it, and the rows on an inset panel with a line between them.
- A table in a message stands out of the bubble, between the text before and after it, and its columns are as wide as their content.
- A table has Copy as Markdown and Copy as CSV under it, as on desktop.
- A code block looks as a table does: the language and the copy button on a card, and the code on an inset panel.
Fixed
- A tap on a chat in the home list now always opens the chat. Before, after the app opened from a notification or connected to a server, a tap on a chat could do nothing until you restarted the app.
- A tap on a notification now opens its chat. Before, it could stop on the home list.
- An attachment download from a server that the server stops, or that arrives damaged, now fails at once. Before, it waited one minute for a timeout.
- Leaving a chat while dictation stops no longer leaves the voice action waiting.
- The back button of a chat always goes back to the chat list. Before, after a swipe back that you started and did not finish, the back button did nothing until you opened the chat again.
- When the selected server does not answer at start, the app no longer stops on the "Connecting to server" screen for up to 30 seconds. After 5 seconds the home shows the server as offline, and you can open the drawer or select another server while the app continues to connect.
- Chat and channel list previews, awaiting replies and search show a pasted file by its name, not as
@image.png. - When ElevenLabs refuses your key or a voice, the app now shows the reason that ElevenLabs gives, such as a free plan that does not allow the voice. Before, it showed "ElevenLabs did not answer", also when ElevenLabs answered.
- When you open a chat again, the messages that came while you were away show at once. They no longer play word by word with haptics, as if they were new. Only a reply that comes while the chat is open plays.
- The text of your messages is white on a dark message color. Before, the text was always dark and was hard to read on a dark agent color.
- After dictation ends, one press of Send sends the message. Before, after an earlier message, Send could do nothing until you touched the text field. This occurred mostly on Android, and sometimes on iPhone.
- In a voice conversation, the voice no longer reads the start of an agent's thinking before the answer.
- Search shows the agents whose name matches first. Before, an agent that had the word only in its description or last message could show above it, and the agent you named could be hidden in All.
- A connection that the app restarted many times, such as after many returns to the app, no longer drops when you open the app.
- In a message with a message color, code blocks, inline code, tables and mention chips follow the text color of the bubble: darker on a dark color with white text, lighter on a light color with dark text. The code has the color of the message text. Before, they stayed grey on the color, and code could be white on a light color.
- A code block in an agent's reply has a background that you can see. Before, it had the same grey as the bubble.
- When the first server in your list does not connect at start, such as a computer that is turned off, the app opens your first online server again. Before, a server that the phone kept from the last visit stayed open, and its chats did not load.
- Dictation in Chinese, Japanese or Thai no longer puts a space between two parts of what you said.
- The thinking sheet of a long run opens at once. Before, a run of many steps made the app stop or close when you opened its thinking.
- The thinking sheet shows all the steps of a long run, also after the app starts again. It reads the earlier steps from the server when you scroll up. Before, it showed only the newest steps.