Changelog

Right-click an agent, a stage or a group to remove it, and a selected connection can be deleted from its inspector. Boards used to be add-only unless you knew the Delete key worked on a focused canvas.

A stage board says so on the board now, and converts to connections in one click. That is how it gets loops, because a loop lives on a connection. The sheet lists what changes before anything changes, including what a debate stage loses.

It used to sit greyed out with no explanation. Now it says to pick two or more agents, shows how many you have selected, and reads as guidance rather than an error.

Three animations were redrawing the window at your display's full refresh rate, including one that changed three times a second and one that ran with nothing happening. They tick as often as they need to now.

The Runs tab groups by workspace, so repeated agent names across repos are finally tellable apart. The most urgent run still comes first.

Draw a connection backwards while holding Option and it becomes a loop. Give it a round limit, and a reviewer to satisfy, and the run repeats until the work passes or the rounds run out.

A connection can now be a review, a revision or a delegation, not just a handoff. Each one briefs the next agent differently, so a reviewer critiques instead of continuing.

A step check can return a number out of 100 instead of pass or fail, and a loop can wait for that score to clear a bar you set.

Add an output node and the run writes its results to a path you choose, so a board can say what it is actually for.

Every task now says its state in a sentence. "Ticked complete, but nexus's branch has 2 commits that never landed" tells you what to do, where a badge made you work it out.

Turn on the Plan view skill and any plan an agent writes opens as a live page beside the work, re-rendering as steps get ticked.

The plan page asks git, not the plan. A task ticked complete whose branch never reached the base branch is drawn as unlanded, in red, next to whoever claimed it.

Settings ▸ Data ▸ Full backup saves your memories, workflow boards, run history, assistant conversations, agent sessions and workspaces into one file. Restoring on another Mac repoints every path for you.

It saved theme, keybindings and agent profiles, never your memories. It is now called "Settings file", so nothing calls itself a backup that is not one.

Settling a paused workflow used to hide it while it still held its board, so every later run of that board was refused by something you could no longer see. A run that still holds something now stays visible.

Right-click a workspace and choose Relocate. Memories, workflow boards, assistant history and agent sessions all follow the folder to its new home. If anything cannot move, nothing does.

A workflow that stopped for you now appears the moment it pauses, instead of waiting for something else to redraw the panel.

Eleven flat rows became six, with the eight colours in one submenu and a tick on the current one. Reveal in Finder and Copy Path are there too.

Finished runs under "recent" can be forgotten one at a time, or all at once from the section header. Until now the list only ever grew.

Pane widths are measured in terminal columns now, not in toolbar buttons, so an agent's output stops wrapping into confetti. A pane never falls below 40 columns.

When a row runs out of room, panes move to a new row underneath instead of scrolling sideways. Every agent stays on screen, because a pane you have to scroll to is a pane you stop watching.

Every run row now carries a visible control to settle it. Marking a run as seen and undoing either choice stay on the right-click menu.

When a step fails, the run pauses there and keeps everything already finished. Fix the task and continue; work that already succeeded is never repeated.

Give a step a check (a file that must exist, a command that must pass) and it counts as done only when the check passes. A run can also carry a contract: what it must deliver, and what must never break.

Merging while an agent works in your main checkout would rewrite files underneath it, so Solenta refuses and names the agent. Two merges into one repo now queue instead of colliding, and a branch switch under running agents gets reported instead of passing unnoticed.

Mark a run as seen, or settle it once it is over. Anything that never landed, whether an unmerged branch or an open pull request, stays visible under "not landed yet" alongside your open PRs and their checks.

Every agent now arrives knowing three things: land your work and confirm it landed, look at visual changes before claiming them, and never say tests pass without the command and its exit code.

A pane never shrinks below the width its controls need. When a row runs out of space it scrolls instead of squeezing, so the tools in every pane header stay reachable.

A new Runs tab shows each orchestration, workflow and lone agent: who is working, who is asking you something, what it cost, and what it produced.

Settings now offers Stable or Nightly. Nightly builds arrive as soon as they pass their tests, and switching applies on the next check.

Give a run a name when you start it. The name titles its tab and appears on every merge confirmation, so you always know which run is asking.

Splitting the same direction now gives every pane an equal share instead of halving the last one, so headers and their controls stay reachable.

An isolated memory could drift far enough off-canvas to drag the whole star map out of view. Stars stay put, and session bookkeeping no longer crowds the picture.

Workflow edges now bow into gentle curves, with arrowheads that follow the flow into each node.

Drag a connection near a node and it snaps to the rim, with a clear "linked" hint when that link already exists.

After a run, every connection shows what happened: green where output was delivered, red where an agent failed, glowing while flow is arriving.

Each node carries a badge: queued, running with elapsed time, done with duration, or failed.

Click a traveled connection to read the exact text that moved downstream, then jump straight to the pane that produced it.

Connections always draw where they belong, show their direction with arrowheads, and can be deleted with a click.

On a board without stages, Run follows your connectors: agents run in order and hand their results downstream.

The traveling pulse now lights up along every connection while a run is under way.

Workflow and orchestration agents keep their auto-approve setting, even after an app restart resumes them.

Arm the orchestrator, reviewers, and each worker with their own skills, right from the Orchestrate sheet.

An orchestration remembers its setup on the tab, so you can re-run or tweak it in one click.

Finished worker panes now clean themselves up, and the orchestrator reclaims the space.

Agent notifications now use a softer sound.

Design multi-agent workflows on a visual canvas. Open it from the new top-bar button.

Pick a reviewer and a PR reviewer, set each agent's model, and finished workers close themselves.

Agents start ready to work without approval prompts, land in one tab, and the orchestrator reacts the moment a worker finishes.

Review and tweak the generated prompt before you kick off a run.

Import skills from a file, a zip, or a GitHub repo, plus nine curated built-ins.

Set the model when you launch from the palette, split a pane, or an agent launches another.

See what changed after each update, and a changelog now lives on the website too.

A new welcome walks you through setup and installs your agents in one click.

Choose and preview a voice during setup, from on-device to cloud.

Drop files, folders, or images into a pane to hand them straight to an agent.

The voice orb now glows with your assistant instead of jittering.

Flip a skill on or off and running agents pick it up without a restart.

The Orchestrate panel lets you choose the model each agent runs.

The menu-bar overlay sits exactly in the notch and stays out of your way.

The orb pulses to your assistant's real voice level.

Solenta now updates itself, so you always have the latest fixes.

A glowing orb shows when your assistant is listening and talking.

A status overlay in the notch shows what your agents are doing while you work elsewhere.

Kimi and Grok join Claude and Codex as assistant options.

Kick off a coordinated multi-agent task from one place.