What changed in Fleetly
Latest releases first. To use each one, see the docs.
- Free plan: after the 14-day trial, the account stays free on one machine, to see the agents, get the notifications and approve requests from your phone. Replying, sending work, deliveries and teams are part of Personal.
- More in the vault and the client link: Word, Excel and PowerPoint are now copied (update Fleetly on the computer); the client opens the running app from the link; Deliveries opens from Settings, also on a phone; and roles can be shared with the building.
- Building recipes: a path that repeats becomes the house way, with a name, and newcomers use it with Do it this way. In the Command center, "how do we do X here?" answers with the recipe.
- Offboarding with handover: when you remove someone from the building, Fleetly shows what they delivered and what’s open under their name (teams, spots, computers, tasks) and hands it all over to whoever you choose. The deliveries stay in the building.
- Do it again this way: each building delivery keeps how it was done (the request and the tools, in order, reviewed by you when approving). Whoever joins opens the delivery and makes a new one the same way, with their own accounts.
- Building deliveries: on Team and Enterprise, what is approved on the delivery card or goes out through the outbox is kept in the building, without depending on the computer of whoever did it. Only the result, never the conversation. In Deliveries, on the top bar, you search by project, label, type and period and download the files.
- Outside clients approve by link: in the Deliveries tab, Send to the client creates a link for someone without an account to see the delivery on their phone and tap Approve or Ask for a change. The reply reaches your phone; a change request only goes to the agent when you send it. No conversation, code or computer on the page.
- Second opinion on the delivery card: another agent on the same computer (Claude Code or Codex) reads what changed and says whether it found a problem, before you approve. It only reads. Redo with the points sends the points back to the agent that did the work, and in the project you can have the review requested automatically when the agent finishes.
- Teams that are easier to follow: a REJECTED goes back only to whoever did it (before, it also moved forward); deliveries open full screen, with formatted Markdown, images and SVG; the roles New team creates belong only to that team, and the library keeps the ones you reuse; while the team runs, the board is view-only; and the top of the board flags the connector you still need to connect (a posts team’s Instagram, for example).
- Talk by voice: on the main screens, the round microphone in the corner opens full-screen listening, with Bit. Speak freely, with pauses: Fleetly recognizes the project by name (“in NFCe, …”) and opens the chat there, or a simple chat. The agent comes from Settings (Sonnet is suggested) or from what you ask (“ask Codex…”). Then it becomes a conversation: Bit thinks, reads the answer aloud and listens again; “go ahead” puts the same chat to work, with the strong model. On the iPhone, dictation now joins the sentences after pauses.
- Google Cloud sign-in from the panel: when the agent asks you to “run
gcloud auth login” and stops, you can sign in through Fleetly, from your phone too: the Sign in button shows up in the chat and under Accounts on this machine, for the account and for the application credential. You authorize on Google’s page and paste the code.
- Bring tasks from GitHub, Linear and Jira into the Plan: in the Project view, Bring from outside. The task shows the number from there, and finishing it in Fleetly comments (and, if you want, closes) it there.
- Reminders on a session: “notify me if it isn’t done by 6pm”, “when done, notify me” and “when done, send: run the tests”, in the Details tab. Works with the panel closed.
- Check when done: up to 3 commands per project (e.g.
npm test) that the machine runs when the agent finishes after changing files; the result shows on the delivery card. - Stuck session notice: your phone gets a notice when a session goes 40 minutes without saying anything, repeats the same error or does the same action in a loop.
- Governance for companies (Enterprise): in the building policy, session rules (maximum sign-in duration, only from certain networks and a recent sign-in to change what matters), retention of up to 10 years and legal hold, Opus only with approval (a model outside the list asks for approval instead of being blocked) and Fix automatically (the app rewrites hooks that were removed). Emergency access now asks for a reason and notifies the other owners. And the audit log can go to the company’s own bucket (S3, JSONL), in Reports → Integrations.
- Set up an agent, no terminal: computer connected with no agent? The panel shows a card for each one (Claude Code, Codex, Antigravity and GitHub Copilot) with Install and Sign in on this computer, which opens the sign-in in its browser. Away from the computer, I’m away from the computer signs in with a code, from your phone. Also under Set up another agent… in New conversation.
- Your agents’ week, by email: every Monday morning, Bit sends last week’s numbers (deliveries, approved, agent time and cost, compared with the week before), what moved in each project, routine and team, and a tip about a feature you haven’t used yet. Turn it off in Settings → Account → Emails.
- Claude meetings in sessions: with Claude’s Team or Enterprise plan, Claude sessions read the meetings Claude joined (summary, action items and transcript). Ask “bring today’s meeting action items into the Plan”. Allow it once per machine in Settings → Connected services → Claude meetings.
- Always allow on connectors: when a connector writes without sending anything to anyone (revealing a contact in Apollo, creating a record, changing a field), it no longer goes to the Outbox; it’s a regular request, with Always allow and Accept all in this session. Emails, messages, comments and posts still always wait for you. And signing in to connectors like Apollo, which rejected the permission list, now works.
- Fleetly browser: agents open web systems with no connector (a CRM, an ERP, a portal) and update data for you, in a browser just for Fleetly on your machine, separate from yours. Turn it on in Settings → Connected services. Reading goes straight through; saving, sending or deleting waits for your approval. If the system asks for a sign-in, you sign in through the conversation’s Live browser, with the page beside it (full screen on a phone), tapping and typing, from your phone too.
- Your Claude account connectors in Fleetly: in Settings → Connected services, each machine shows the connectors you already turned on at claude.ai (Gmail, Drive, Notion…), and Claude sessions use them with nothing stored in Fleetly. Connect on claude.ai opens the catalog to add more with one tap. In Claude roles and routines, Account connectors picks which ones to use: Read only ones are read without asking for approval, so an overnight routine reads Gmail without stopping, and anything that writes goes through the Outbox. Codex shows the apps connected in the ChatGPT account, and the Get started proposal says which connectors are already connected and sets up the roles with them.
- Fleetly apps in the stores: the Android app is on Google Play and the iPhone and iPad app is on the App Store. Follow your sessions, approve requests right from the notification (with Face ID or fingerprint) and open a preview of what the agent is building. Sign in with the account you already use. For now, only in the Brazilian stores.
- A start for non-developers: the first time you open Fleetly, it asks what you work with (sales, marketing, legal, management, software) and opens on What do you want the agents to do?, with examples from your field told in 4 steps (the request, who does it, the progress and the delivery to approve). Connect my computer explains the app and the agent subscription (Claude, ChatGPT with Codex, Gemini with Antigravity or GitHub Copilot) in plain words and waits for the computer to show up; a request written earlier is kept. Outside software, the dashboard starts without the developer features (you can turn them on in Settings). The getting-started guide is now about results: connect, ask, turn on notifications and approve the first delivery. And the Report summary runs with any agent on the machine, not just Claude. For non-developers, the 1-minute tour, the tips and the help videos speak the same language too, with new videos on connecting the computer, asking for the first job and approving a delivery.
- Waiting for you, in Today: at the top of Today, what needs your decision (permission requests, sends, sessions ready for you to review and team rounds to publish), each opening in the right place. The session-ready notification is only sent when it changed project files (saving Claude’s memory doesn’t count).
- Delivery card: when the agent finishes, the Deliveries tab opens with everything waiting for your judgment: what it said, the app running right inside the card, what changed in the code and the delivered files. Approve commits only what this round changed; Redo sends your request, typed or dictated, straight to the agent. The phone notification sent when the session finishes after changing files opens right on the card.
- Today, so you don’t get lost between chats: the Today button in the tab strip (or the h key) lists the day’s chats by the time of your last request, with what you asked in quotes, since the automatic name changes during the day. In the expanded panel it becomes a column on the right; on the phone, full screen. Tabs also no longer disappear when the collector restarts. And Dismiss now holds until the session has something genuinely new (it used to come back on its own) and also removes the session from the phone widget and lock screen.
- Clearer notifications and a widget with a little bot: each notification says the project and what happened, with a symbol (✋ request, 📋 plan, ✅ your turn, ⚠️ team stopped), how long the session worked and what it replied. Permission requests no longer arrive twice (one came without buttons), plans have Approve and Deny too, and notifications and buttons follow the device language. In the 1.1 apps (on the way to the stores), a home screen widget with a little bot that types while agents work, raises its hand when someone needs you and sleeps when all is quiet; and progress on the lock screen (a Live Activity on iPhone, a pinned notification on Android) with how many are working, who needs you and the round clock.
- Fairer reviews in Teams: the reviewer only rejects what another seat can fix; what only you can check (your phone, a system app) goes to the list of things to check. And a seat on a machine where the team branch is already in the main folder no longer gets confused by the branchless copy. Routines now also run with Antigravity and Copilot, created and edited right in the list. The Teams help videos were re-recorded.
- Quick fix in Teams: a small fix goes straight to a single seat (the frontend, for example), skipping the PO, the plan and the others, with a review afterwards if you want. The report card can be minimized, the summary shows a skeleton while it is written, and the Teams header and the Routines screen are tidier.
- Clearer teams at the end and along the way: the round report starts with For you, in plain words (what is done, what to do now and what is not tested yet), for people who aren’t technical. While the team works, the top bar and the tab title show the team, and the board shows how long each seat has had the work, how long it usually takes and its latest actions. And the team keeps track of what gets left behind: a seat’s pending work carries on to the next, tests only against a simulation become a point to check, and integrations with an official format (like the invoice XML) are validated against the spec.
- No-tunnel preview on your own computer: with the dashboard open on the same machine as the app, Open preview goes straight to
localhost. And the preview through the hub opens again (it sometimes said “preview not found”). - Routines that build a whole feature: describe the feature and the routine works in three steps in a worktree: the agent builds it, the other agent reviews it (Codex if Claude built it, and vice versa) and the first one fixes it. Routines now run with Claude or Codex, also trigger on every commit to a branch or when a session finishes, can open the pull request at the end (the agent still can’t push) and send a single “Rotina terminou” (routine finished) notice.
- Usage leftovers, one tap: instead of running on its own, the machine can just suggest on your phone the routine for the leftovers; tapping the notice opens Fleetly with Run. And the Report shows the usage put to work from leftovers.
- A retro written by an agent: turn on An agent writes the week’s text and the weekly retro gets a few paragraphs on what moved forward in each project; Write the text asks for it on any saved week. A building owner can turn the retro on for everyone in the building, on the chosen day and time. In the Office, The week on the meeting room’s sign opens the retro, and the ▦ on your room’s sign opens the Board.
- Clearer when connecting a computer: from the app, when the machine is already in another building, the page shows which one it leaves and which it joins, and only switches after you confirm.
- Notices for Slack and Teams: the important integration events now include a permission request waiting for 15 minutes and a machine going offline (and coming back).
- Antigravity subagents: they show up inside the conversation that opened them, with what each one did.
- Antigravity in the terminal: when
agyasks whether it can run a command, the session shows up as waiting for you, with the command, and the notice reaches your phone. With the Antigravity hook (in the collector installer), the building policy also applies to those sessions and, with away mode, you approve from the dashboard. On Windows, the dashboard now sees which sessions are working. - New chat, like Claude Desktop: the Nova conversa button (or the n key) opens an empty tab with the box already focused: just type and send. The project, agent, model and what it can do are chips around the box, set to what you used last; picking a project makes the same screen open a development session, with a worktree.
- Plan with an empty sheet: the period summary looks at the projects your agents worked on in those days.
- No browser dialogs: confirmations in the Plan and in Delete account now use Fleetly’s own dialogs.
- Zoom in the Plan: the − 100% + in the bar zooms only the sheet, and so does a trackpad pinch or two fingers on the phone.
- Tooltips: hovering over a button shows a Fleetly tooltip, in the light or dark theme, instead of the browser’s.
- Ask the agent, in the Plan: describe a feature and the agent you choose reads the code and returns the plan and the feature broken into tasks, in order, which you accept into a new list in one tap; or ask it to look at the code and suggest what needs attention. It runs in a Plan only session you can follow and continue.
- The Plan, one more step: the team queue (while the team is in a round, the task waits and starts on its own) and the report’s verdict on the task; the period’s summary can come out on its own on the first day; notifications use your time zone; and in a sprint edited together with the building, teammates run it too.
- The Plan with teams and notifications: Send to a team starts a round with the task and it comes back as your turn at the end; Plan the project summarizes the whole project; Numbers shows how long tasks spend with the agent and waiting for you; a sprint can be edited together with the building; the Plan and your phone tell you when the period or sprint ends and when the summary is ready; and tasks show up in the project list and on the Board.
- Periods and projects in the Plan: the Periods view is your planning across projects, one period per sheet (a week, two weeks, a month: + Next period suggests the name and dates and carries the open tasks you choose). The Project view organizes each project its own way: Sprints (dates and a goal, with a review at the end), Periods (blocks of time without a goal) or Lists (no dates), with the Backlog, Pull into your period and a summary of each.
- Plan the week in the Plan: an agent on your machine (Claude Code, Codex, Copilot or Antigravity, whichever you pick) summarizes what got done and what is waiting for you in each project, lays out the week’s or sprint’s focus and suggests tasks you accept one by one. Summarize project does the same for one column. And a free column ("Sprint 12") now holds tasks from several projects, which go to an agent right from there.
- Approve the plan: in Plan only mode, Claude’s plan arrives in a card you can read: Approve plan and it carries it out in the mode you choose, or Ask for changes and it reworks it. With the other agents, Plan ready? below the conversation. And in Teams, Plan first: the first seat writes the round’s plan and the team only starts once you approve.
- Help for the agents: in ⚙ Team, files (the architecture, how to test, the project rules; Word and PDF too) that reach every seat in the team before each handoff, without going into git.
- Publish all the way: in ⚙ Team → Publish, the team can also merge the pull request and run your deploy command, with the output on screen, in a single tap. And the report shows when an agent asked you for something its role does not allow.
- Team round report: at the end of every round, what actually happened: the verdict, what to check, the commits and files read from git (kept apart from what the agents said), the preview on the right screen and the next steps. The phone notification carries the numbers.
- Contact us: from your menu (or Settings → Account), send a Suggestion, a Bug or an Issue to the people behind Fleetly, with a message number and, if you want, the screen and versions included (never session content). My messages shows the status and the reply, which also arrives by email and on your phone. See Your account.
- Team results: the new Teams tab shows everything the team did across rounds: deliveries, approved and rejected, commits, lines, how long each seat worked and what was done.
- Teams leave your folder alone: the team’s code lives in its own copy of the folder; yours stays on the branch you are on. And Approve everything in this team, or per seat, so the team doesn’t stop asking for approval.
- Plan: a sheet to plan your week or sprint, in the same handwriting and paper as the Board. One column per project (or free columns, like "Week 45"), with colors, tasks by type and Send to an agent, which opens the session in the project folder. The task comes back with your turn when the session ends, and you are the one who checks it off. Only you see your tasks, unless you show one to the building.
- Everything shows up right away: the dashboard gets changes from sessions, the Office and the radio as they happen, instead of asking every few seconds. If the connection drops, it goes back to asking on its own.
- Teams on a phone: the board fills the screen, and Roles and Round slide up from the bottom, one at a time. Tapping a seat or an arrow shows its details.
- Arrows with approval in Teams: on an arrow, choose who approves before it moves on (you or people in the team’s building; with several, one or all). The work stops there, and the approvers get what was delivered on their phone and in Teams, to Approve, Ask for changes (it goes back to the seat you choose on the arrow, like a flowchart) or Decline. For a PO to analyze the idea with you before anyone builds it, or a marketing text to go through someone before it goes out. See Agent teams.
- Cleaner top bar: the filters left the strip over the stage and moved to the Filters button, next to the views, which opens a sheet with what to show, the period, the project, the machine and the tag. Active filters show in the corner of the stage, with × to remove them. New session stands out in the bar, and Command, Report, Teams and Board sit together in one group. Under Fleetly, next to the machine, you see the building you are in at the Office.
- Lighter Office: the agents’ floor uses much less CPU (from 69% to about 10% in our test, with a full floor sitting still): only the rooms on screen are drawn, each at its own pace (fast only where an agent is typing), and the whole floor no longer has to fit in a single drawing. Companies with dozens of people, who saw a blank floor before, now see everything.
- Light mode in the Office: the Light button, at the bottom right, stops everything that moves; rooms only change when a session changes state. Good for keeping the panel open all day on a laptop.
- Full rooms: in a room with more than 12 conversations, the ones that leave the view are the least recent (they go to the archive), never the ones that just arrived or finished, and no desk changes place.
- GitHub Copilot in the dashboard: GitHub Copilot CLI (
copilot) sessions show up next to Claude Code, Codex and Antigravity, with the Copilot badge, including Copilot chats in VS Code, where the dashboard’s message lands as if you typed it and permission requests reach the dashboard and your phone (experimental). You can chat with them, start new ones in + (on Copilot Auto or with a model from the list), approve each command and each edit from the dashboard and your phone, put Copilot in a role or on a team seat, and hand a conversation off to it or from it to another agent. Install it withnpm install -g @github/copilotand sign in with/login(or through the machine’s existinggh). See GitHub Copilot. - Inventory with every agent: in Settings → Reports → Inventory (and in the CSV and JSON export), each machine also shows its Antigravity and GitHub Copilot versions, next to Claude Code and Codex.
- Team stuck without quota: when an agent runs out of quota (or fails with an error) mid-round, the team stops and tells you in the log and on your phone, instead of waiting for an answer forever. In Antigravity, the error also shows up in the session’s conversation.
- Run the project on an Antigravity seat: a short session of the machine’s Claude starts the app, without asking for approval on every command and without touching the seat’s conversation. And Antigravity’s own internal ports no longer show up as an app Running in the folder.
- Approvals that take a while: a permission request from sessions started by the dashboard (Claude, Codex and Antigravity) was denied on its own after 5 minutes without an answer, even with Wait for an answer up to set longer. Now it waits for the time you chose (15 minutes in Antigravity). And in Antigravity’s Auto,
git -C folder statusand reads chained with&&(cd folder && git status && git diff) also run without asking, likegit status. Update the machine’s collector. - Antigravity fix: an Antigravity session started with a role (including on a team seat) showed up in the role’s internal folder instead of the project’s, so the team couldn’t carry that seat’s code through git. Update the machine’s collector.
- Quick tips: in the bottom corner of the panel, one tip at a time with what isn’t obvious from the screen, like when to use a chat or a team, how to schedule an agent and everyday shortcuts, with Learn more for the docs. × hides them; Settings → Chat brings them back.
- Each team seat’s agent: each seat chooses which agent it runs in (Claude, Codex or Antigravity), under Agent, on the seat or in the People step of New team, without changing the role on other teams. A team template that asks for an agent your machine doesn’t have already comes with one it has, and on an existing team the top of the board offers Use Claude in this seat. Bug triage and Risk mapper now suggest Antigravity (if you don’t have it, they stay on Claude).
- A machine with nothing installed: the Fleetly installer and app show what the machine has and what’s missing (Claude Code, Codex, Antigravity and git, each with the official command for the system) and offer to install it. In the app, it’s under Agentes neste computador, with Instalar and how to sign in to each one. See Connect machines.
- Agent fixes: an interrupted Codex session (or one that hit an error) no longer stays marked as working for up to half an hour, refusing messages; and a team no longer stops when the agent writes DONE: at the end of its last sentence instead of on its own line. The Windows install command (PowerShell) works again: it stopped with "illegal characters in path".
- Antigravity in the dashboard: sessions from Antigravity CLI (
agy), Google’s terminal agent that replaced Gemini CLI for personal accounts, show up next to Claude Code’s and Codex’s. You can chat with them, start new sessions from + (with your account’s model: Gemini, Claude or GPT-OSS) and approve from the dashboard and your phone, with Accept automatically and the building policy in effect. It also joins roles and teams next to Claude and Codex (for example, Claude writes and Antigravity reviews), and you can hand a conversation off to it and from it to the others. See Antigravity. - Approvals for sessions started from the dashboard: requests reach the dashboard and your phone again for Claude sessions in Ask for approval (they were denied right away) and for the Codex hook, where the building policy applies again. Update the collector on your machines.
- Time in the chat: each of your messages shows when it was sent, and each answer when it finished, including sessions from VS Code and the terminal.
- More roles and teams: development gets Architect, Database, End-to-end QA (uses the app in the browser like a person would), Interface reviewer and Technical docs, plus the New screen template. The recipes add big change, database migration, email sequence, cash flow and roles for fact checking, lead qualification, meeting prep, collections and onboarding plans. The blog post now checks every number against its source before the brand review.
- Rename a building: the owner renames it in Settings → Buildings, next to the name.
- Buildings in one place: in Settings → Buildings, each building has Machines, People, Plan and Security. Each machine is now a row (online or not, system, sessions and a warning when it matters) that opens the rest when tapped, and Add machine asks one thing only: with the app or with a command. The plan shows who pays and each building’s usage. If you have a single building, you go straight into it. See Buildings and people.
- Ready-made teams by area: in New team, the templates now cover development (Full feature, with PO, backend, frontend and reviewer), marketing, sales, support, finance and operations, each role with its name and prompt ready. The recipes add security review, page SEO, one piece for every channel, sales proposal, help article and open a job. See Teams.
- Governance for companies (Enterprise): managed protection puts Fleetly’s hooks in Claude Code’s managed settings (
sudo fleetly-collector hooks --gerenciado, or through MDM), and whoever uses the machine can’t remove them; the Inventory shows each machine’s protection, the building gets an alert when it drops, and the policy can require a minimum protection (below it, the machine is read-only). The audit log now has proof outside the database: every day a signed copy goes to locked storage, and the integrity attestation checks whether anything was touched. In Reports → Integrations, send the audit log and the agent trail to Splunk, Datadog or a signed webhook, or pull them through the API with a read-only token. - Share the week: in Report → Weeks, each week now has Share, which builds a card with what your agents did, to post on LinkedIn or X. You choose what shows; project names start off, and nothing leaves your browser until you download, copy or post it. See Tools.
- Recipes on the site: the Recipes page (in the Who it’s for menu) shows Fleetly's ready-made teams, roles and routines, by area (dev, marketing, sales, support, finance and operations), with open prompts and the steps to use them in the dashboard.
- Privacy in buildings with several people: on each machine’s card (Settings → Machines), whoever looks after it picks what teammates in the building can do: as their role allows, read only (they see the sessions but can’t send messages, approve, commit or push) or can’t see the sessions. And on Team, keys and passwords in conversations are now masked for everyone, as on Enterprise.
- Every routine and pages for each audience: in Teams, Routines shows the routines of every project and machine in one list, with Run now. The guide counts every track (not just the first). And the site got For developers and For teams pages.
- Demo building and per-screen tours: if you haven’t connected a machine yet, tap See a demo to see the dashboard at work, with sample sessions, an agent team mid-round, the Board and the Report, with nothing to install. And the Report, the Board and the expanded session got a 1-minute tour on your first visit. See Getting started.
- Videos and help: in your menu, every short video in one place, with a 3-minute full demo (from the first agent to the team that publishes on its own), and a ▶ on screens that have a video. The guide got a Go further track (team, routine, Board and retro), and your first visit asks whether you’ll use it on your own or with your team, so the tour shows what matters. See Getting started.
- Clearer menus: Teams and Board now have labels in the top bar, next to the views. Today, Summary, the projects, the weeks and the people live together in the Report, in tabs. And, so they don’t get mixed up with agent teams, the per-person numbers are in the People tab, and the Organizations tab in Settings is now Buildings. See Tools.
- Agent questions in Fleetly: when Claude asks a question with options before going on, it shows up in the dashboard and on your phone as Question from the agent, with the options and a field for another answer, and the answers go back to it as if you had picked them in the tab. And Accept all in this session no longer answers these questions on its own (before, they went through with no answer). For sessions in VS Code, close and reopen the tab to get the new bridge. See Approvals.
- Teams: publish from your phone, first request and the walking folder: at the end of the round, the notification comes with Approve and Deny: approving opens the pull request from the team branch to the main branch (with team code through git) or commits and pushes the team folder’s branch, and merging stays with you. The New team wizard got a First request, which can start the round right away. And in the Office with Rooms by team, each handoff carries a folder from one desk to the other. See Teams.
- Weekly retro: turn it on in Settings → Notifications and approvals and choose the day and the time (default: Friday, 6 pm). Once a week you get a notification with what the agents delivered in the projects you see: sessions, time and commits compared with the previous week, the projects that delivered the most, the routines and the teams. The weeks are kept in Report → Weeks. By project only, never by person. See Tools.
- The app’s collector updates itself: the collector inside the Fleetly app (Mac and Windows) now follows the hub like the one installed by the command: within 30 minutes of each release, with the signature checked and without waiting for a new app version. If the new version doesn’t stay up, the app goes back to the one it shipped with. Starts with app 0.1.22. See Connect machines.
- Command: which machine it runs on: under the title, Runs on shows the machine whose Claude replies, preferably one of yours; with more than one, pick from the list. Before, someone operating several machines could send the message to one and see another one’s chat. See Command.
- A more compact chat: the session status moves up to the title line, tags sit at the end of the details line, each agent action becomes one line (tap to see the command and output), and at the bottom shortcuts, previews and the long-chat notice share a single line. The text is a bit smaller, with A− and A+ to adjust it. See Chatting with sessions.
- Team code right on the screen, and your teammate live: in Teams’ Code tab, the file opens alongside, without leaving the team: yours, to edit and send with Send to the team (commit and push to the team branch; the team gets a notice and the code reaches idle seats right away, or through Get new code); a teammate’s, view only. And when a teammate has the file open in Fleetly’s editor, you see their line and what they type before saving. Showing is each person’s choice, only for the team, and nothing is stored. See Teams.
- Codex on Windows working for real: Fleetly now turns on Codex’s Windows sandbox (the one that doesn’t need administrator rights). Without it, Codex refused every command, even
git status, and a Reviewer on Windows rejected work without being able to read anything. Needs app 0.1.21. And in Teams’ Run the project, the preview link shows up as soon as the app starts, without reloading the page. See Teams. - Clearer Teams: the Teams screen now follows the team’s path. At the top of the board, what’s left before running (Build → Prepare → Run), and tapping it takes you there; on the right, the selected seat with Now on top; the team’s request below the board, with Round finished to ask for more (the same sessions remember what they did) or Start from scratch; and name, folder, code, copy and delete in ⚙ Team. The seat that’s working pulses and shows what it’s doing now, and the arrow the work goes through moves. Help answers each question with a short video, every heading has a ?, and there’s a screen tour. See Teams.
- Teams with outside people, without friction: accepting an invite became a three-step card (where the agent works, the team’s code and show to the team), and whoever works on their own laptop, in their own building, shares just the team folder with the company’s building, from anywhere and from the phone: the team sees that folder’s sessions, and the rest of the machine stays theirs only. Private GitHub repositories get access on their own, and the Code tab shows the whole team’s code. See Teams.
- One team, one codebase: with Team code through git, every seat works on the team branch, each on its own machine; Fleetly brings in what the others sent before each seat starts and commits and pushes when it finishes, and whoever accepts an invite gets the repository already cloned. In the editor, a bar tells you when another seat is changing the same file, with See what it changed before the push. And New team became a four-step wizard (template, code, people and review). See Teams.
- Teams with people: on Team and Enterprise, a team can belong to the building, and each seat can belong to a different person, including people working from home or at another company. They get the invite in the panel, on their phone and by email, accept by choosing their machine and folder, and that seat’s agent runs there, with their account. Building invites can go by email too. See Teams.
- Teams on the Board: each team becomes a card on the Board, with its last round and every seat’s state; inside it, all seats, the log and the team’s notes, and one more level for each seat’s session and what it delivered. See Views.
- Nothing left running: the apps Fleetly starts with Run the project stop on their own after 2 hours (you can extend it), even with the dashboard closed, and the top bar shows how many are running, with the list to see and stop them. See Teams.
- Stop the app, and who runs in each seat: next to a running app’s address, Stop ends only what runs on that port, in the session’s folder. And each team seat shows who runs in it (Claude or Codex, from the role), with the session list limited to the right agent. See Teams.
- Run the project with any agent: in a Codex seat, a short Codex session in the same folder starts the app, with internet turned on in its sandbox, and the app keeps running after the answer (in Claude too). Codex roles got the Internet in Codex’s sandbox option. And a seat shows when its session is from a different agent than the role’s, with Replace for the right one. See Teams.
- Teams, one more round: a seat receiving from several waits for all and gets it all together, and the round only ends when no one has the work; Chats shows every seat’s chat side by side; copy the team to a new folder; carry the code through git between machines, with a guide; and, at the end, Run the project with the preview. The dashboard’s push now also creates the branch on the remote when it’s new. See Teams.
- Teams that are easier to steer: Hand off unsticks a seat that finished without DONE; each seat shows where it stands; a seat can work on another machine (Claude on the server, Codex on the Mac); Copy to… reuses the team in another project; a session shows which team it belongs to, and closing a session opened from the team takes you back; and See in the Office opens rooms by team. See Teams.
- Teams in the Office and imported agents: in the Office, Rooms by team shows each team as a room with its seats’ sessions. In Roles, import the agents you already have in Claude Code (
.claude/agents/*.md). And when the team finishes, the notification opens the team with Review and publish. On machines with the Fleetly app, importing agents comes with app 0.1.14. See Teams. - Arrows with conditions and team templates: an arrow can apply only if APPROVED, if REJECTED (back to whoever built it, with a loop limit) or on another word, like UI or API. And in New team, start from a template: One builds, the other reviews, Product team or Bug duty, with Claude and Codex on the same team. See Teams.
- Teams that work on their own: write what the team should do and tap Start. The request goes to the first seat and, when each seat finishes with
DONE:, what it delivered moves along the arrows to the next one, even from Claude to Codex and across machines, with the Team log showing each step. And sending from the dashboard to a Codex conversation no longer fails when the Codex extension in VS Code is slow to answer. See Teams. - Teams: the org chart button in the top bar builds your agent team from your roles: each seat shows, live, the session in it, opens a session with the role already set or takes an existing session, and arrows show how the work flows. A team can have Claude in one seat and Codex in another. See Teams.
- Roles: in New session → Role, create roles for your team (frontend specialist, reviewer, triage…), each with the role prompt, skills, agent (Claude or Codex), model and what is off limits. A session opened with a role follows it in every message, and its name shows in the header. It is the start of agent teams. On machines with the Fleetly app, it comes with app 0.1.14. See Roles.
- Claude and Codex on the same work: in a Claude session, Hand off to Codex (in the message box’s +) takes the work to the other agent, and in a Codex one, Hand off to Claude. The agent that will continue reads the conversation and writes the handoff, so it works even when the first one’s usage window is used up; the new session opens in the same folder and branch, with both agents’ project rules (
CLAUDE.mdandAGENTS.md), for you to review before sending. On machines with the Fleetly app, it comes with app 0.1.14. See Chat. - Board: all your projects on a whiteboard, live, over the whole screen or in a window of its own (board button in the top bar), each with the machine, folder and building it comes from: stage, what was done, what the agents are doing now, what is waiting for you and the week. The same repository on another machine is the same project, and with the tag filter on it opens on the tag board. And your notes: ideas and reminders on sticky notes only you see; a reminder reaches your phone and an idea goes to an agent. Claude writes each project summary (where it stands, next steps and decisions) when a session finishes a step; a project can have tags that new sessions get; a project note can be shown to the building or made into a routine; and the whiteboard in the Office rooms opens the project board. On machines with the Fleetly app, grouping the same repository, the summary and the project tags come with app 0.1.12. See Views.
- Routines: in the project list, create tasks that run on their own on a schedule, in a new session: review what changed, run the tests and fix, or whatever you write. They can run in a worktree and accept automatically, with a time limit; push and pull requests are always left to you. And usage window leftovers: close to the reset, with usage left over, the machine runs a marked routine instead of letting the limit go to waste. On machines with the Fleetly app, it comes with app 0.1.11, which also removes from the menu the new collector version notice when there is nothing to install. See Routines.
- Morning brief: in Settings → Notifications and approvals, one notification a day, at your time, with what is waiting for you and what the routines and sessions did since the night before. See Approvals and notifications.
- Go to a session: search (
/) understands project and topic together, likea1 deploy, shows the matching sessions right away and, with nothing typed, the ones you opened last. Each session shows when it did a deploy, push and commit, and the project list has the Last deploy at the top. On machines with the Fleetly app, the marks come with app 0.1.10. See Tools. - Chat as a timeline: like in VS Code, each step of the chat (your message, answer, thinking, action) has a dot, joined by a line from start to end. See Chat.
- Project list: tap the project name in a session’s header (or the room’s sign) to see all its sessions with a summary of each, and filter by name, summary and chat text. The chat search shows up to 5 matches per session; on machines with the Fleetly app, that comes with app 0.1.9. See Views.
- New session with Codex, any way you start it: the Claude/Codex choice shows whenever some connected machine has each one, and picking one the machine lacks switches to a machine that has it.
- Accept automatically: in a session, Accept all in this session approves every permission request in it right away, until you turn it off; the building’s policy still applies. Each approval is in the agent trail, in Reports, with who turned it on. Requests also show Approving… and Approved right away, Approve all answers several at once, and the VS Code bridge offers the same options as the tab (such as accepting all edits). App 0.1.8. See Approvals.
- claude.ai artifacts in the chat: when the agent publishes an artifact (a screen preview, a design), the chat shows an Open artifact button, and your phone gets a notification. See Chat.
- Files and editor in the session: the Files tab shows the project’s files and opens an editor, on a phone too. On a computer, with the session expanded, files, editor and chat sit in columns as wide as you like. Save and ask to test saves and writes the request in the chat; if the agent changes the file first, nothing is lost. On machines with the Fleetly app, it needs app 0.1.7 (Install version 0.1.7, in the menu). See Files and editor.
- Remembered permissions: in the session Details, the "Always allow" rules for this session and the project, with Forget on each. See Approvals.
- Preview that reloads on its own: when the agent changes a file, the phone preview updates in about a second. And when it finishes a change with the app running, the notification says Ready to test and opens the session with the preview highlighted. App 0.1.6.
- Newest Claude Code on the machine: with more than one install (npm, native, the VS Code extension’s), Fleetly uses the highest version. If it is still too old for the model, the message says how to update. App 0.1.5.
- Plans: upgrading to Team applies once the first Team payment is confirmed, and anything over the plan limits becomes read-only. See Subscription.
- Permissions in both places: with the VS Code bridge, a permission request shows up in the tab and in Fleetly (and on your phone) at the same time; the first answer wins. Works with remote VS Code too. App 0.1.4. See Chat.
- Security: the collector only updates to versions we signed and never goes back to an older one; account recovery links, the command center, the local permission hook, pairing by code and switching buildings are now more restricted. App 0.1.3 for Mac and Windows with the same fixes.
- VS Code bridge: a message sent from the dashboard reaches the session open in VS Code as yours, same as typing in the tab (nothing gets closed, and the agent accepts deploy, commit and so on). No extension: the collector writes one option of the Claude Code extension. See Chat.
- Send as me for sessions open in VS Code: Fleetly closes the session there and sends the message as your request. See Chat.
- Fleetly app for Mac and Windows: in Settings → Machines, download the app, open it and sign in with your account: you confirm in the browser and pick the building, with no token to copy. It lives in the menu bar (or the tray), shows plan usage and approval requests, and turns on the hooks without a terminal. See Getting started.
- The app updates from its menu: when there is a new version, Instalar a versão … (install version …) shows up, with a signed package. If you have 0.1.0, install 0.1.1 once from the download.
- Lock a single machine (Enterprise): company laptops stay in the building, and a contractor’s can serve others. And the app’s confirmation page warns beforehand when the computer is already in another building.
- More than one building on the same computer: for people who work for their company and a client, for example. Each building sees only its own folders and has its own approval mode. In the app, under Configurar… → O que o prédio vê → Adicionar outro prédio; in the terminal, with
--adicionar. See Connect machines. - Leave the building from the machine itself: from the app menu, or with
fleetly-collector sairin the terminal. See Connect machines. - Spend fewer tokens, in Claude and Codex: Fleetly warns you when the cache of a large conversation has expired (the next message reprocesses everything at full price) and offers Start over with a summary, a new session with a summary of what was done. See Chat.
- Approve Codex in VS Code from your phone with no setup: with the conversation open in the extension, permission requests reach the dashboard, the menu bar and your phone. See Approvals.
- The dashboard now also in English: it follows the browser’s language, and you can switch in Settings → Account.
- Worktrees: a new session can work in a copy of the repository, on a branch of its own, and then open a PR, merge or discard from the Code tab. See Code and Git.
- Isometric pixel-art office, with the illustrated look and the front view as options, and Compact desks mode.
- Automatic approval from your phone: in Automatic mode, requests come to the dashboard and your phone when you are away from the computer. See Approvals.
- Cost and budget (Enterprise): estimated and billed cost, end-of-month forecast and a budget with a lock.
- Full docs on the site, in Portuguese and English, with search.
- One machine, one building: running another building’s command on a machine now asks before switching, and the old building finds out (token revoked, phone notification and audit log). On Enterprise, you can lock your machines. See Connect machines.
- A guide for getting started: your menu (with your initials) has Getting started with short videos, the 1-minute tour, help and Sign out. Fleetly notices each finished step on its own. See Getting started.
- Reorganized settings: tabs grouped (Agents, Building, Your account), with Notifications and approvals together. On the phone, they open as a list, like the iPhone Settings app.
- Codex in VS Code: conversations from the Codex extension show up in the dashboard, and a message you send from Fleetly goes into the conversation open in VS Code. See Chatting with sessions.
- Expanded building policy (Enterprise): folders, network, protected branches, maximum mode, hours, models, MCP servers, minimum version and simulation before saving. See Enterprise.
- Company SSO, with a verified domain, required SSO and SCIM (Enterprise).
- Signed integrity attestation of the audit log, machine inventory, and commits and pull requests in the agent trail.
- Sign in with Google, Microsoft or Apple, and account recovery by email.
- Command blocking on the Team plan too.
- An open dashboard updates itself after a new release.
- The building policy applies before every action, in auto mode and in Codex too.
- Agent trail and security report (Enterprise): every agent action, with who authorized it.
- Executive, access and access review reports, with email delivery (Enterprise).
- Auditor role, "Who saw what" and the "What Fleetly records about you" section in Settings → Account.
- Chained audit log that shows whether any record was changed.
- Personal, Team and Enterprise plans, with a 14-day free trial. See Pricing.
- Folders per machine: the building sees only the folders the machine chooses. See Folders.
- Team view, per person (Team and Enterprise).
- Enterprise: secrets masked at the source, two-person approval and an exportable audit log.
- Collector on Windows.
- Timeline, Strategy and Ambient views; the Office becomes the first view.
- Codex in Fleetly: Codex sessions in the same dashboard, with chat, new sessions, queue and plan limits. See Codex.
- New session in any folder on the machine, in a new folder or as a standalone chat.
- Switch the model in sessions Fleetly drives.
- Preview on your phone of the app the agent started on localhost, with no tunnel and ports found automatically. See Preview.
- Take over in Fleetly and Continue in the Claude app for sessions open in VS Code. See Chat.
- Approvals from the dashboard and your phone in sessions Fleetly drives, and Always allow in this session or in the project.
- Chat tabs, like in VS Code.
- One-line install, and the collector updates itself.
- Accounts with passkeys, buildings, roles, invites and an audit log.
- Dashboard reworked for phones.