Everything that shipped.
Agent releases roll out automatically the moment they’re signed. Here’s what changed, newest first.
Every version, on record.
The canvas tells the story, not just the steps.
The Session Canvas stops being a list of tool calls and tells each turn’s story, from your prompt to how it ended. This month’s agent update also brings September’s last agent-side features, Remote Screen sound and Clu driving a tab among them, to your machines.
6 changes
The canvas tells the story, not just the steps.
The Session Canvas stops being a list of tool calls and tells each turn’s story, from your prompt to how it ended. This month’s agent update also brings September’s last agent-side features, Remote Screen sound and Clu driving a tab among them, to your machines.
Session Canvas story cards: each turn now reads as your prompt, Clu’s plan, the approvals it waited on, test results, sub-agent lanes and how the turn ended. Delete a message and its cards go with it.
The canvas pane in Clu Chat lays itself out around the work: a test loop sets the run beside the edit, the plan shows as a row of step marks, and a finished turn reads as finished at once.
One busy machine can no longer use up your whole fleet’s cross-device command allowance: limits are now counted per requesting machine.
The first voice memo after a few quiet minutes no longer fails.
A sign-in that can’t finish now tells you why, instead of dropping you on the home page.
The dashboard your agent serves locally reconnects on its own after the agent restarts.
Clu gets hands, and shows its work.
Your machines start asking each other’s Clu for help, Clu can drive a dashboard tab you hand it, and an agent’s session reads as cards instead of scrollback. Remote Screen gains sound, and newcomers get a real first-run setup.
12 changes
Clu gets hands, and shows its work.
Your machines start asking each other’s Clu for help, Clu can drive a dashboard tab you hand it, and an agent’s session reads as cards instead of scrollback. Remote Screen gains sound, and newcomers get a real first-run setup.
The Session Canvas: an agent’s session as timestamped cards of every tool call, live inside its Clu Chat thread, in plain language with the exact command one hover away.
Your machines talk to each other: Clu on one machine can ask the Clu on another, which answers from its own memory and tools, and Mesh Talk shows every exchange as one feed.
Let Clu drive this tab: switch it on in Clu Chat and Clu can read and type into your terminals, arrange your windows and click through the dashboard, in that one tab only.
Remote Screen with sound: hear the host’s system audio once you approve it, never its microphone. Click-and-drag and double-click now land, and ⌘V pastes text from the computer you’re on (Mac and Linux hosts).
Welcome to CluTerm: a first-run setup takes a new account from install to an open desktop, then a Welcome app and a desktop tour show you around.
Smart-glasses support (Even Realities G2): Clu Chat with push-to-talk and spoken replies, ambient notifications, and a live terminal tail on the lens, through a Bluetooth Linux ARM puck running the separate glasses build.
Canvases become apps that keep their state across reloads and send your clicks back to Clu, and a canvas never sits blank: it shows Clu’s latest content or says what it’s waiting for.
Clu Chat queues the prompts you type while a turn runs, talks to whichever machine you pick, and says so when a turn runs out of steps instead of passing it off as finished.
Windows machines do more: Clu Chat runs Claude Code inside WSL with Clu’s own tools, cross-device commands run in PowerShell (or WSL on request), and new terminals open in your home folder.
Terminals outlive agent updates: on macOS and Linux the built-in session manager upgrades in place without closing a shell. The agent also stops installing a tmux that shadowed yours, and marks a stuck program “Not responding”.
Connected now: the Devices window shows who is on each machine and over what link, gives each machine an icon of its own, and lets you approve a quarantined machine with its sync password from any browser where your root machine is unlocked.
The Clu Cell: every agent Clu starts runs in an OS sandbox drawn around its project, with your SSH keys and AWS, Kubernetes, Docker and GitHub credentials out of sight (macOS; Linux when bubblewrap is installed, otherwise unsandboxed with a logged warning).
One dashboard for every machine you own.
The dashboard stops being a window into one computer and becomes the Rig: a single mesh view of every machine you’ve paired, with a 30-day free trial to try it on your own fleet. Clu Chat, meanwhile, grows a memory and a voice.
9 changes
One dashboard for every machine you own.
The dashboard stops being a window into one computer and becomes the Rig: a single mesh view of every machine you’ve paired, with a 30-day free trial to try it on your own fleet. Clu Chat, meanwhile, grows a memory and a voice.
The Rig: open the dashboard and see every paired machine as one connected desktop, not a picker.
Multi-screen remote control: enumerate every display on a shared machine and aim your input at the one you want.
A Meta Quest client: step into the Rig in passthrough VR and work alongside Clu in the room.
30-day free trial with simple device-based pricing, no card required to start.
Sleep cycles: Clu quietly summarizes what it’s learned about your projects every night, week, and month, browsable in a new Journal.
Voice and images in Clu Chat: push-to-talk and spoken replies, and screenshots pasted straight into the conversation.
Clu Chat picks up mid-tool-call after a reload, renders code and markdown live while streaming, and lets you choose a model and effort per message.
Spaces remember where each machine’s windows live, even after a reload.
Reconnecting a viewer no longer shrinks your terminal grid over time.
What we’re building next.
Updates land automatically.
Run cluterm update, or let the agent pull signed releases on its own.