Vibezeo 0.11.0

Resource governor.See how much CPU and memory each pane uses, freeze a pane until you need it, and cap an agent's memory.

The Ports tab gets a Panes section that lists every pane running on this computer with its CPU and memory, grouped by worktree, and an agent pane's Dashboard card shows the same, such as "CPU 12% · 840 MB". CPU counts one core as 100%, so 250% is two and a half cores busy. Both refresh every 2 seconds while you're looking. "Freeze", in a pane's ⋯ menu or its row in Panes, stops every process in the pane until you choose "Thaw", and the pane's header says "Frozen" meanwhile; closing or restarting a frozen pane thaws it first. Settings › Engines › Memory cap per agent ("Off" by default, or 512 MB to 16 GB) caps each agent pane started after you set it: when a process in the pane goes over, the system ends that process, and the pane says "Stopped: memory cap 512 MB" while its shell keeps running. With systemd older than 253, the whole pane ends instead. With a systemd user session, each agent pane runs in its own systemd user scope, and caps also need that session to be able to limit memory. Without one, the setting says why and stays "Off", and Freeze stops the pane's processes with signals. Settings › Engines › One check at a time, off by default, runs the check commands of Best of N, the review queue and Land worktrees one after another instead of all at once; a Best of N lane waiting its turn shows "Queued". Readings and frozen states aren't saved, and nothing leaves your computer.

Agent-proof secrets.Keep a key's value in your system keychain and put a reference in .env instead, so agents, diffs and worktree copies only ever see the reference.

"Add secret" in Settings › Secrets saves a value of up to 16 KB under a name such as stripe-test; Vibezeo keeps only the name and never shows the value again. "Copy reference" gives you vibezeo-secret://stripe-test to use in the .env, as in API_KEY=vibezeo-secret://stripe-test. "Run with secrets…" in the ⋯ menu of a terminal pane on this computer, or the play button in its header, asks for a command and lists the variables it will get from the .env at the top of the pane's workspace or worktree. A reference to a secret that doesn't exist reads "No secret named stripe-test", and "Run" stays off. Run opens a new terminal pane beside it, named after the command, where the command gets every variable in that .env, with each reference replaced by its value; when the command ends, the variables are unset and a fresh shell takes over, so what you type there next never sees them. In that pane, a secret's value of 4 bytes or more that the command prints shows as [REDACTED], on screen, in the scrollback and in what Vibezeo saves of the pane. Masking catches only the exact value, and only in that pane: a command can still use a value, write it to a file or print it in another form. Agent panes don't get the values in this release; a way for agents to run commands with them, after you approve, comes in a later release. "Delete local data…" in Settings › Privacy removes every secret too.

Multi-repo workspaces.Give one agent a worktree across several repositories that sit side by side in one folder, such as api and web.

Add the folder (it can't be inside a repository itself) and its rail row says how many it holds, such as "2 repositories": it needs 2 to 12 repositories directly inside it, and a repository's own worktrees beside it count once. "New worktree…" and the launch sheet's "Give each agent its own worktree" then make one folder inside it, .tandem/worktrees/<name>, holding a worktree of every repository, all on the branch tandem/<name>, and the agent starts there, so it sees them all. If one repository can't get a worktree (one with no commits, say), none is made and Vibezeo says which. "Open diff" shows each repository's share, such as "api · 3 files", with every path under its repository's name, and your comments reach the agent as for any worktree. "Create pull requests" asks first, then in each repository with new commits pushes the branch with your own git, opens a pull request with your GitHub CLI (gh) and links it to the others in its description; repositories without new commits, and GitLab ones, are skipped and listed. "Discard worktree…" removes every repository's worktree and the folder, and keeps the branches. Merging, landing, the overlap check, pull request checks, previews, worktree setup and second opinions don't work for these worktrees yet: their menu entries are off, under "Not available for multi-repo worktrees yet", and the intent radar skips them. Folders on SSH hosts are never multi-repo.

Any ACP agent in Thread mode.Run any coding agent that speaks the Agent Client Protocol (ACP) as a thread.

Gemini CLI, through gemini --acp (or --experimental-acp on older releases), and Kiro CLI, through kiro-cli acp in each account's own home, now run threads too, beside Cursor Agent, Grok and OpenCode. To add another, "Add an ACP agent…" under Settings › Engines › ACP agents takes a name, a command (on your PATH, or its full path), arguments and environment variables, and a Goose preset fills in goose acp. "Test" starts the command, asks it to introduce itself, waits up to 10 seconds and stops it, then shows its name, version and how it signs in; a command that doesn't answer as an ACP agent is refused with the reason, and "Add" runs the same test before it saves. The agent then appears in the Thread engine picker labelled "via ACP", and runs from the path saved when you added it, on this computer only. You can add up to 16. Each agent signs in its own way, and Vibezeo never sees its keys: a variable whose value looks like a key or token is refused, and when an agent asks for a sign-in, the thread names the ways it offers. "Remove" stops the agent's threads; they stay readable but don't run again, and an agent you add later, even under the same name, is a new engine.

Claims.Agents in worktrees of one repository say which files they're about to change, so you see two of them heading for the same file before git reports a conflict.

Agents with Vibezeo's tasks server get three tools: claim takes up to 50 files or folders at a time (a folder covers what's in it, and a pattern isn't allowed) with a note on what the agent is changing, release gives them up, and list_claims shows what agents in the repository's other worktrees claimed. A worktree's briefing asks its agent to claim before it edits. Claude Code in a terminal pane also claims a file just by editing it: only the file's path leaves its hook, never the change. When two worktrees claim the same file, or one claims a folder holding a file the other claims, both worktrees' agent panes show one line, such as "add-oauth and fix-auth both claim src/auth/session.ts", even a pane in a tab you open later, and the Dashboard's review queue lists every overlap above its items. A worktree's claims go when it's merged, landed or removed, and the main checkout's when its last agent session ends. What one agent wrote in a claim reaches the others marked as data, never as instructions.

Plan with Lead.Give Lead a goal bigger than one task, and it splits it into tasks, each with the engine it suggests, for you to edit and approve.

"Plan with Lead…" on the Tasks page takes a goal of up to 4,000 characters and a workspace on this computer. Lead plans in a new chat there, at Read access, where you can follow its reasoning and answer it; it has no task tools in that chat, so nothing reaches your tasks before you approve. It sees the installed engines that can run tasks and how much of each account's usage window is used. When Lead waits on you, the plan says "Lead is waiting for your answer in its thread", and after 30 minutes of work Vibezeo stops it and the plan says why. Lead's reply becomes a list of 1 to 12 tasks you can edit: their titles, descriptions and engines, or remove one. Characters that can hide text are taken out. Creating the tasks puts them in a new project named after the goal, as To do and written as Lead, each remembering its engine, which Execute… then starts with. Execute all, which names how many tasks it runs (such as "Execute all 3…"), first lists each run: the task, its engine and that engine's account with the most room left, its own worktree, and the engine that reviews it, with the tasks it leaves alone and why. Nothing starts until you confirm; then each task runs once, in its own worktree, at most 4 at a time, and finished work shows in the review queue. Merging stays with you. A plan stays in the Tasks page's header until you choose "Remove plan", which keeps its tasks, or quit Vibezeo.