What we’ve
been shipping.

Everything here already shipped.

Newest: Task titles are written in your words
Ship log · Jun 17 to Sep 1068 updates
Jun
Jul
Aug
Sep
38 shipping days. One block per update; click one to jump to it.

September 2026

12 updates
Improvement

Task titles are written in your words

Agents were asked to name tasks in everyday words, but only the questions they send you were actually checked, so titles like "Build the competitive search baseline" and "Diagnose the first retention opportunity" still reached the board. The board now holds every agent-written task title to the same check as those questions, whether the task is created from a task, routed by the lead, or started from Chat: a title that uses words like evidence, baseline, diagnose or audit is sent back to the agent to rewrite before anything is created. Titles you type yourself are never checked. Existing tasks keep their names.

Open Tasks
Fix

Chat shows what each agent step touched again

The activity rows under an agent's reply had stopped naming their target: a website read showed no address, a saved document no title, and a page fetched from the shell appeared as "Running command". The rows read their target from the arguments the agent runtime reports, and a presentation rewrite had dropped the format the runtime actually sends. Targets are back: the site and path that was read, the document title, the file name, and the specialist profile name written as words rather than an identifier. Grouped page reads show the site plus how many more pages were fetched.

Open Chat
Fix

Specialists close their first task the way the board expects

A specialist finishing its first task was reading one set of rules for how to close it while the board enforced another, so a close that looked correct to your agent was refused and the attempt was spent finding that out. Both now describe the same thing: the specialist proposes its recurring job for your approval and closes on that, or closes on the published document if you decline it.

Open Tasks
Improvement

Resetting a workspace keeps your connected providers

Resetting a workspace used to throw away every model provider you had connected, so before you could start again you had to paste each API key back in and sign in to Codex from scratch. That was the longest part of a reset, and worst if you had several credentials set up to rotate. Your providers now survive a reset the way your integrations already did: the provider step shows them still connected, your rotation order and fallback chain come back with them, and the rebuilt workspace configures itself without asking you for anything. The reset dialog now lists what is kept alongside what is deleted. Everything else a reset removes is unchanged, and you can still remove a credential yourself in Settings.

Open Settings
Fix

Setup says what actually went wrong

When a squad launch was refused, the reason was being lost on the way back to your lead, so it retried the same plan several times without changing anything and then told you setup had failed. The reason now reaches your lead, which can correct the plan and carry on. If a failure genuinely has no explanation behind it, setup says so plainly instead of showing you an internal diagnostic.

Open Chat
Fix

A reset workspace can build its squad again

A workspace that had been through Reset workspace could finish the whole setup conversation and then refuse to launch, reporting a persistent internal setup failure however many times you or your lead tried again. The reset keeps the name you go by, so your lead never asked for it a second time, and setup went on waiting for an answer it already had. Setup now reads what you have actually told it rather than only what arrived in the last message, and any workspace stuck in that state launches without redoing the conversation.

Open Chat
Fix

Setup stops offering a retry that cannot work

Setup used to answer every failure the same way: we couldn't bring your workspace online, please try again. Some of those failures are ours, and repeating the attempt returns the identical answer, so the retry button spent your time on a result it could not change. Setup now separates the two. A failure that another attempt could clear still offers the retry; one that needs a fix on our side says so and points you at support instead of at a button. Behind that, deploying SquadRoom now restarts the service that configures workspaces, so the code we ship is the code your setup runs against.

Open Settings
Fix

A restored workspace starts its container again

Restoring a backup stops the workspace to apply it, and the workspace used to go on advertising the container that stop had removed. Setup then ran against an address with nothing behind it and reported that it couldn't bring the workspace online. A stopped workspace now says it has no container, and setup rebuilds one before doing anything else.

Open Settings
Fix

Resetting a workspace says it is resetting

Resetting a workspace used to drop you on the "Setup needs another try" screen while the reset was still running, because clearing the container looks the same from the outside as a workspace that failed to start. Retrying from that screen rebuilt the old workspace and left the reset unfinished. A reset now holds the workspace for the whole job: you see a reset in progress until it lands you back at setup, nothing else can restart the workspace underneath it, and a reset interrupted by a closed tab can be picked up where it stopped.

Open Settings
Feature

Request changes to a Doc from the Doc itself

Every Doc now has a Request changes button in its header. Say what should change and SquadRoom routes it for you: the Task that saved the Doc's current version reopens for its agent when it is finished, receives your note as a comment when it is still running, and a new Task is created for the lead to assign when no Task is left to take it. The Doc is tagged in the request automatically, and the line above Send tells you which of those will happen before you send.

Open Docs

August 2026

32 updates
Improvement

A redrawn light theme across the whole workspace

Light mode now runs on one set of surfaces, borders and action colours instead of page-by-page variations, so the board, Squad, Docs, Calendar and settings sit in the same frame. Scrolling happens at the window edge rather than inside nested panels, selected text is a pale wash instead of a solid bar, and the Docs space switcher collapses into a dropdown on narrow screens.

Improvement

Image thumbnails on task attachments

Files attached to a task now show an image thumbnail or a file type badge instead of a plain link, and images inside a document render in place when you open it. Screenshots your agents deliver are visible without downloading anything.

Open Tasks
Feature

Flag anything an agent wrote

Feedback is a note to the SquadRoom team: it never reaches your squad and changes no Task, Report or Ticket. A flag now sits on every agent comment in a task, alongside chat messages, reports and deliverables, and carries the item and a short excerpt with it, so a single tap is enough to tell us a result was wrong. The user menu holds a general entry for everything else.

Improvement

Cheaper browser sessions and earlier credit warnings

A browser session now costs 10 credits. Your workspace also warns you once when the balance drops to 15% and once when it runs out, rather than leaving agents to meet the limit mid-task. The usage page pages through longer task lists and shows which agent spent what.

Open Settings
Feature

Suggestions get their own view

Work your squad suggests now sits beside Board and Tickets as a queue you move through one card at a time: Start queues the work, Dismiss moves to the next. The dashboard's "Needs you" count is now Tickets only, the work genuinely paused on your answer, with suggestions reduced to a single pointer row underneath.

Open Tasks
Improvement

Fewer suggestions, each one worth reading

Agents were closing almost every task with a suggestion attached. The bar is now written where they read it: propose only work you would bet the owner starts, never routine follow-through or scope already owed, and closing with none is a normal outcome. Work an agent still owes you comes back as a Ticket instead.

Improvement

Setup chat gets to the point

The squad setup conversation no longer ends every turn with the same closing question or draws each recommendation as the same arrow chain. When you answer the naming question, it builds the squad in that same turn instead of waiting for another nudge.

Improvement

The playbook catalog, rebuilt around execution

Weekly Pulse folds into Week Plan: score last week, then plan this one, in a single document and a single job. Social Pulse joins as a daily scan that drafts only when something real turns up. Publishing Loop, Search Radar and Market Watch now end in a deliverable or a drafted response rather than a memo, Support Desk counts recurring questions over time, and Morning Brief may say nothing needs you today. Each loop keeps one document it versions on every run, so its history reads as one chain.

Open Playbooks
Improvement

Scheduled job approvals read like a decision

The approval card now leads with the prompt in plain words, how often it repeats written as words instead of cron, the company number the loop is meant to move, and whether it keeps a learning document. The long agent instructions fold behind a toggle. Schedules run on your workspace clock, so nothing has to stop and ask you for a timezone.

Open Calendar
Improvement

The board answers "is anything running right now"

Columns are drawn from work actually in flight, so a task paused on your Ticket or waiting on a child no longer sits in In Progress with nothing happening. The drawer marks a running task LIVE, and a parent card names the child task blocking its next step.

Open Board
Announcement

Icebreak, a free cold email generator

Icebreak writes cold email openers from what is publicly known about a prospect. It is free, needs no account, and sits in the site navigation next to Gauntlet Loop.

Feature

See what each agent is doing right now, on the board

Every In Progress card now shows a green pulsing dot and a short line under the title that says what the agent is doing at this moment, in plain words: "Pulling the 3 overdue invoices from the books", not a tool name or a log line. Agents update it as they move between steps, the board updates live, and the line disappears the moment nothing is running. No click needed to see where the work stands.

Open Tasks
Improvement

Every playbook now ends in something you can act on

Playbook runs no longer stop at a summary. Each installed playbook now defines the exact output it produces and how the run ends: an Approval or Decision Ask with the recommended option marked, up to three Proposals for work it cannot start, or a stated reason nothing needed doing. Weekly Pulse keeps a living scorecard against your Company Goal, Support Desk and Inbox Triage bundle their drafts into one send approval, Pipeline Builder lets you pick prospects from the shortlist before any draft is written, and Market Watch rates each change by impact before recommending a response.

Open Playbooks
Feature

Four new playbooks: Week Plan, Unblock Sweep, Ship Log, Customer Voice

Week Plan opens the week with up to five Tasks, each with an owner and a finish line, as one Decision you approve. Unblock Sweep finds work that stopped moving, posts the unblock where an agent is stuck, and bundles close or reassign verdicts into one Decision. Ship Log turns finished Tasks into a changelog entry and a customer-facing release note draft. Customer Voice counts what customers keep saying and proposes one change for the loudest theme.

Open Playbooks
Feature

Every task has an activity timeline

Tasks now carry a structured timeline of what happened and when, in order, with agent mentions linked inline. You can follow a task from kickoff to delivery without opening raw logs.

Open Board
Feature

Rename an agent

Give an agent a different name whenever you want. Earlier names stay registered as aliases, so older tasks, reports and chat references still resolve to the same agent.

Open Squad
Feature

Your lead delegates, specialists propose

The squad lead can hand a piece of work to the right specialist rather than carrying it alone, and specialists can put forward work they think is worth doing. Proposals still wait for your verdict before anything starts.

Open Board
Improvement

Skills are now Playbooks

What used to be called Skills is now called Playbooks throughout the app, matching what they are: a repeatable piece of work your squad runs. Links and chat references can point at one playbook and open it directly.

Open Playbooks
Feature

A month view for the Calendar

The Calendar has a month view you can page through, so a whole month of scheduled work is visible at once. A single job can hold several schedules, and past runs keep a history you can open from the day they ran.

Open Calendar
Feature

A notification inbox

Mentions and new Tickets collect in a notification inbox in the app, so what is waiting on you sits in one place. In-app notifications are limited to things addressed to you rather than everything your squad does.

Open Dashboard
Feature

Watch your squad work

Agent activity in chat is an inspectable timeline. Each step an agent takes, including the pages it reads on the web, appears as it happens and stays there after the run, so you can open a step and see what was actually done instead of reading a summary after the fact.

Open Chat
Feature

Start a skill from a template

A library of guided templates covers common recurring work, and setup collects the workspace details a template needs before the skill is installed. You pick a starting point instead of writing a skill from a blank page.

Open Skills
Feature

Every agent has its own chat channel

Chat is organised into a channel per agent, with threads that persist between visits, so a conversation with one specialist stays separate from the rest of the squad. Empty threads say so instead of looking broken.

Open Chat
Feature

Tag your tasks

Tasks take tags, so you can group work by project, client or theme and find it again on the Board.

Open Board
Feature

Set up your squad in a conversation

Squad setup happens in chat. You describe what is eating your time, the lead asks follow-up questions, and it proposes a squad for you to confirm before anything is created, instead of asking you to fill in a form.

July 2026

23 updates
Feature

Whiteboards in chat

Your squad can sketch on a whiteboard and the result appears in chat as a canvas you can open and edit, which suits a flow or a layout better than a paragraph describing it.

Open Chat
Feature

Light mode

SquadRoom has a light theme alongside the dark one, with surfaces, dividers and status colours tuned for it rather than inverted.

Feature

Task proposals wait for your verdict

Recommended scope now lands as an inert proposal instead of starting on its own. Your squad lead screens the slate first, and the surviving proposals wait in Tickets until you approve or decline them.

Open Board
Improvement

Scheduled jobs appear on the Board

Scheduled jobs run as visible Tasks with the same lifecycle as everything else on the Board, so each recurring job's progress and output are tracked beside ad hoc work.

Open Board
Feature

Reference docs and tasks inline

Point at a document or another task from inside chat and task prompts. Pick the reference from a menu as you type and your agents pick up the linked context.

Open Chat
Feature

Browser screenshots in chat and tasks

When an agent browses the web as part of its work, it can capture screenshots that show up inline in chat and on tasks, with full-size viewing.

Feature

Attach files and images in chat

You can now upload files and images directly in chat with your squad. Images show inline previews, and clicking one opens a full lightbox gallery so you can flip through everything shared in a conversation.

Open Chat
Feature

Built-in user guide

New in-app user guide with navigable sections. Look up how any part of SquadRoom works without leaving your workspace.

Feature

Image generation for your agents

Agents can now generate images as part of their work. Configure image generation in workspace settings; generated media is served through secure, workspace-scoped routes.

Open Settings
Feature

Workspace activity log

A new paginated activity log shows what has been happening in your workspace, loading more as you scroll.

Feature

Agent personas with distinct voices

Each agent on your squad now writes in its own voice, and onboarding builds a researched roster tailored to your goals — so replies read like a team, not one bot with different names.

Improvement

Mobile-friendly interface

The app now works properly on phones and small screens: board, chat, and settings all adapt to smaller layouts.

Feature

Billing history

Your billing history is now tracked and viewable in settings, alongside easier access to the billing portal.

Open Settings
Feature

Workspace backup and restore

Back up your whole workspace and restore it later, with server-side storage. A reset flow lets you start a workspace fresh when you need to.

June 2026

5 updates
Feature

App integrations

Connect external apps to your workspace with OAuth sign-in, and let your agents use those tools as part of their tasks.

Open Integrations
Feature

OpenRouter model support

You can now bring your OpenRouter key. Provider setup moved into a dedicated modal with key validation, so misconfigured keys are caught before your agents need them.

Feature

Talk to your squad on Telegram

Connect a Telegram bot during onboarding and message your squad straight from Telegram — task updates and replies flow both ways.

Announcement

SquadRoom — first release

The first cut of SquadRoom: a task board for your agent squad, agent memory, and a scheduled-tasks calendar. Everything since builds on this foundation.