Week of 14 September 202612 changes
We split onboarding into a quick account connection step and a first task, brought the workspace to iPhone and iPad, and made pages behave like the rest of your tabs. We also tightened the handoff between you, your Agent, and the Computer where the work happens.
Start with a real task
We split the first visit into two steps. Connect an account such as Gmail, Calendar, Drive, Notion, or Linear, then choose a starting point and tell your Agent what you need. New tasks use your Space’s default model, so you can begin without configuring the model first.
The first-task guide now floats over the conversation instead of taking space out of it. It tells you whether your Agent is working, waiting for you, or finished, keeps saved drafts recoverable, and lets you ask for a change or mark the result useful.
Runva fits in your hand
We shipped a native iPhone and iPad app that keeps the same workspace and remote Computers as the web app. Threads, Activity, Computers, sharing, and the rest of the workspace are one tap away.
The composer stays above the keyboard, Return makes a new line on touch devices, and remote browser pages respond to taps, scrolling, and the system keyboard. On iPad, the wider layout stays open when the window allows it.
Pages are part of the conversation
Pages now sit in the same tab strip as Threads, Automations, files, and browser tabs. You can drag them, move through them with the keyboard, and use the same close and context-menu actions.
In a Room, switching the resident Agent switches to that Agent’s own tabs and workspace view, so the person you picked stays the one you’re looking at.
Getting started
- We replaced the shared signup code with invite links. Members get five links at Invites, and each person who joins gets a Personal Space.
- We let you restart onboarding after deleting the first Agent, so a cleared first attempt doesn’t leave you without a way back in.
- We show first-task failures in the conversation with a path forward, and preserve an unsent draft when sending hasn’t been confirmed.
Agents and Computers
- We added a Computer picker to Agent settings. You can attach an existing linked Computer, create a private managed Computer, or leave the Agent without one and use its Connections instead.
- We split Agent settings into separate sections for profile, approvals, run behavior, and ownership, so changing one concern doesn’t make the others harder to find.
- We added model release dates and sorted models by capability within each price band, so the model picker is easier to compare.
Connections and Docs
- We made connection setup recover on its own after you sign in. During onboarding, the status refreshes while the provider return is finishing, and the flow tells you whether an account is connected or your organization has disabled it.
- We gave desktop OAuth returns their own handoff back to Runva, and added a clear result page that can show what happened even before you sign in again.
- We made Docs keep their files with them. An Agent can read a Doc’s files and files from other pages, and save a Doc with files attached.
Week of 7 September 202610 changes
We added shared replies in Rooms, made the workspace easier to use on phones and tablets, and simplified creating an Agent. You can start with a message and organize the work as you go.
Threaded replies in Rooms, too
You could already start a reply Thread in an Agent DM. Now you can do the same in a Room, keeping a discussion about one message separate from the main conversation.
Everyone in the Room can take part. The original message stays at the top, and you can find the discussion again in Threads.
More comfortable on a smaller screen
We made buttons and menus easier to tap, kept the message box above the keyboard, and made settings fit a phone. On touch screens, Return adds a new line; you use the Send button when the message is ready.
Your phone remembers its own sidebar preference, so closing it there won’t change your desktop layout. On an iPad, you can keep the wider workspace columns open.
Make an Agent by sending the first message
Choose New agent and write what you need. Runva creates the Agent when you send that first message. You don’t have to pick a name or fill out a setup form first.
Your workspace
- We added custom sidebar sections. You can group Agents and Rooms together, rename and reorder your sections, and choose a sort order for each. On a phone, use Move to section in the row menu.
- We added Search in room to the Room menu, so you can find a message within that conversation.
- We fixed message menus getting clipped at the top of the transcript. You can reach the actions on the first visible message.
Connections
- We put available setups first in the Connections directory. For providers you can’t connect yet, we explain what’s missing and link to the provider’s requirements.
- We added support for custom API-key headers when connecting an MCP server, so you can use services that need a different header format.
Also
- We corrected the context meter to use the latest model step’s reported input, with estimated growth since then. Previously, it could substantially overstate how much context you were using.
- We redesigned the website with a new finish-line logo, printed illustrations, and stronger blue and green. You can explore example Agent work on the homepage before signing in.
Week of 31 August 202624 changes
The workspace finally has somewhere to start: sign in and you get a blank Thread, not whichever conversation happened to be first. We also gave Agents the rest of the browser, since clicking and typing only gets you as far as a login box.
The workspace has a home page
Signing in used to drop you into whichever Agent came first in your list. Now you get a home: a new Thread, asking what you want done.
Your Agents are on a dial under the message box. You turn it to pick who gets the message, instead of scrolling a list.
The rest of the browser
An Agent could click and type. It can now scroll, press keys, drag, pick from a dropdown, tick a checkbox, and upload a file.
It also won’t click the wrong thing when the page changed underneath it, or do the same thing twice when a Run gets interrupted.
Sign in without giving an agent your password
When an Agent reaches a password, a card number, or a one-time code, it asks you to fill it in. The value never enters the conversation, and the Agent never sees it.
The prompt reaches you wherever you are, including your phone.
Working with an agent
- We shipped Add folder in the composer’s attach menu of the desktop app, so you can point an Agent at a folder on your own machine.
- We fixed where your message goes when you send it while an Agent is working. It now appears in the transcript at the point you sent it, and the Agent’s reply is split around it. Previously, the reply was one block that grew around your interruption.
- We removed the twelve-step limit on a Run when nothing else sets one. Previously, a Run stopped after twelve passes of tool use and reported itself completed. An Automation’s own step budget still applies.
- We shipped Activity, your Threads grouped by what’s happened in them, so you can catch up on what moved while you were away.
- We removed the limit on how many Agents you can make, on every plan.
- We fixed an Agent blaming somebody for a failed Run. Previously, it could tell a person they were holding up its work, as though the failure were waiting on their approval.
Your computer
- We made taking over a Computer feel like using it. Your typing and dragging keep up with you, and the pointer changes shape the way it would on your own desk.
- We gave cloud Computers a desktop you’ve used before, with a dock and Chrome as the default browser. An update won’t rearrange a desk you’ve set up.
- We put the notes for every release you skipped into the update offer. Previously, it named only the newest one.
- We fixed the browser panel going blank while a Computer sleeps. It keeps showing the last page it had.
- We shipped a way to replace the machine behind a linked Computer without linking it again.
- We added a last-seen time to a local Computer that’s been away more than two days.
The workspace
- We shipped live updates to the workspace, so you won’t need to reload it to see what changed.
- We fixed the command palette highlighting the wrong part of a multi-word query, and stopped the list shuffling under you while you type.
- We fixed the workspace pane following you between conversations, so whether it’s open is remembered per Thread. Previously, opening another Thread on the same Agent inherited the last one’s pane.
- We fixed Back to app in settings, which now returns you to the Thread you left from. Previously, it went to whichever Agent came first in your list.
- We merged plan and billing into one settings page.
- We added a link to every message, so you can point somebody at one.
- We fixed the billing page showing a new plan before the payment was confirmed. It now waits for confirmation when you come back from checkout.
The desktop app
- We added a badge on the app icon with the number of Threads waiting on you.
- We shipped notifications for new messages. Clicking one opens that message in the app you already have open, instead of reloading it.
Week of 24 August 20264 changes
Reopening a conversation you were just in shouldn’t feel like opening it for the first time. This week we fixed that.
A thread you were just in is still there
Come back to an Agent or a Thread within half an hour and it’s there straight away, with the transcript, the activity, and the tool calls you’d already read still in place.
We refresh it when something’s actually changed. Switching to another window and back isn’t a change, and we no longer treat it as one.
Long conversations scroll like short ones
Scroll back through months of messages and the older ones are already there by the time you reach them. No spinner at the top of the transcript, and no jump when the next batch arrives.
Also
- Go back to an Agent, Thread, or Inbox you were in recently and you land on what you were looking at, not on a loading state.
- If we ever fall behind on updates, we reload rather than show you something that might be out of date.
Week of 17 August 202614 changes
The biggest week so far. We renamed Tasks to Threads, gave every new Agent a computer of its own, and stopped showing you a half-drawn workspace.
Tasks are Threads now
By calling a conversation a Task we were promising things we couldn’t keep: an owner, a status, a moment where it’s finished. Most work with an Agent isn’t shaped like that.
A Thread is a conversation with the same Agent, kept separate so it can have its own context. Whether something is running, waiting on you, or scheduled comes from the work itself, so you never have to keep a status up to date.
Every new agent gets a computer
Make an Agent and it comes with a private computer already attached. The machine exists straight away, and we only start the meter once the Agent wakes it.
Agents can turn their own computer on and off, too. We answer a power-on immediately and warm the machine in the background, so the Agent can carry on thinking while it waits. Powering off is a real decision about somebody’s work, so we tell you what it will interrupt before you make it.
One dial for how hard an agent thinks
Low through max, five steps, set per Agent. One setting means the same depth whichever model you’ve chosen, because we translate it on the way out. You never have to learn one provider’s vocabulary to get a considered answer out of another.
Working with an agent
- We do web search ourselves, instead of making you hand an Agent a tool for it.
- Type @ to bring an Agent, Thread, or file into a message. We put it in as text, so you can delete the words to take it back.
- Send something mid-turn and choose whether to steer the run that’s going or queue it for the next one.
- Rate any finished reply with a thumbs up or down. On a thumbs down we ask what it should have done instead, and let you send the rating without answering.
- We photograph a computer when it wakes as well as when it sleeps, so there’s always a current picture to show you.
The workspace
- We wait until the whole workspace is ready before showing it to you, rather than hand you part of one.
- Conversations open on the most recent messages instead of waiting for the whole history.
- Pick a search result and you land on the exact message, in place, with the conversation still around it.
- Take control of a computer and we no longer reload the screen underneath you.
- We now say why an empty pane is empty, including when the computer belongs to somebody else, which isn’t something you can fix by clicking.
- The live view of a computer starts a good deal quicker.
Week of 10 August 20263 changes
Where all of this starts: an Agent you talk to, with a real computer behind it.
An agent with a computer of its own
Chat with an Agent that has files, a browser, a terminal, and a screen you can watch it use. Ask for something and it does the work on a real machine that keeps what it made.
Also
- Navigation is instant. The frame appears the moment you click and the contents fill in behind it.
- We stopped replaying a conversation’s animations every time you go back to it.