Circuitry Changelog

Components version independently: the Desktop app, the EServer (tray / CLI / Docker), and the web editor each have their own version. When a change requires you to update your EServer or MCP to keep working, the note says so and Circuitry will prompt you.

Desktop app

1.4.1 — 2026-08-08

• The top bar can be made shorter. Settings → Appearance → Header Density offers three heights on the desktop app and iPad. The height you have today is the largest of the three, and the buttons keep their full-size icons as the bar shrinks, so they stay easy to find and press.
• The server button shows where your work runs. Grey for this computer, green when another computer or a cloud server is connected, and a cloud when it is a cloud machine. A single cloud machine now shows whether it is awake.
• Fixed: the server menu stayed open when you clicked the canvas.
• Fixed: the desktop app could offer to move your project to a cloud server as though this computer had been lost.

1.4.0 — 2026-08-07

• Your project can follow you between computers. When the connection to your computer drops with a project open, the files panel offers a choice: reconnect, or hand over to your Circuitry cloud server and carry on there. The first time, Circuitry walks you through setting the project up on that machine — open a terminal on it, git clone your repository, choose the folder — and after that, switching between your machines is automatic, with a note telling you which machine you are looking at. Sync your work between locations the way you already do: commit and push before you move, pull when you arrive.
• A lost server connection is noticed in seconds, not minutes. Stopping or quitting your server tells Circuitry immediately, and an unreachable server now reads as a calm "Offline" rather than a red error with advice you can't act on.
• Cloud servers sleep on your schedule. Sleep-when-idle is now a setting (10 minutes, 30 minutes, or 1 hour) — and a busy server no longer dozes off mid-task: real activity on the machine (terminals, files, code runs) keeps it awake, and true idleness lets it sleep so it stops spending your credits.
• Cloud server addresses are one click away. Hover a running cloud server in Settings to copy its server and app web addresses — handy when you're testing something you built on it.
• Cloud servers always come with Python ready, verified at creation.

1.3.0 — 2026-08-05

• The chat and the terminal are two views of one conversation. Switching between the assistant chat and the terminal beside it used to start a second, separate assistant that knew nothing about the first — two histories, neither aware of the other. They are now the same conversation: carry on in the terminal, come back to the chat, and everything you said in either is there.
• Say something while the assistant is working and it hears you now. A message sent mid-task is delivered into the work in progress instead of waiting for the whole task to finish, so you can redirect an assistant that has set off in the wrong direction rather than watching it finish and starting again.
• You can watch the work as it happens. What the assistant is doing appears in the conversation as it goes, in its own block, rather than arriving all at once at the end.
• Attachments land where your cursor is. Attaching an image inserts a marker at the caret, so you can write about a specific picture in the middle of a sentence and the assistant knows which one you mean. Removing one renumbers the rest.
• A connection that drops now says so. If the connection to your server is lost part-way through a reply, the chat says the connection dropped instead of quietly presenting the half it received as the finished answer.
• Assistants can draw joined-up diagrams. Ask for a system diagram and you get properly labelled boxes with arrows that connect them, sized and weighted to what they join — and it can be turned into a working flow with the "To Nodes" button, which previously did nothing on a diagram an assistant had drawn.
• Thick arrows leaving a drawn card render correctly, instead of occasionally drawing a large white oval over the diagram.
• A chat carries on where it left off after a reload. If Circuitry restarts while an assistant is working — a crash, a refresh, or a restart you asked for — the chat now rejoins the work in progress instead of being left with half an answer. Chats that are still working show their spinner in the tab and in the open-chats list, and their replies keep streaming in.
• Nothing you had typed is lost. A half-written message, the files staged beside it, and any messages queued behind a working assistant all come back after a reload, and the conversation reopens where you were reading rather than jumping somewhere else.
• Scrolling up to read no longer pulls you back down. While a reply streams in you can scroll back through the conversation and stay there; the chat resumes following along when you return to the end.
• Terminals come back showing what happened while you were away. A terminal running a coding assistant used to reappear blank after Circuitry restarted, even though the work had carried on. It now reconnects and shows the output it missed, so you can see where the task got to.

1.2.1 — 2026-08-04

• Attach files to any chat, with or without a project folder open. Attaching an image to a chat backed by a coding assistant used to fail with an error unless a project folder was open — now it just works everywhere.
• Word, PowerPoint, Excel and OpenDocument files are now readable in chat. Attach a .docx, .pptx, .xlsx, .odt, .odp or .ods and every model — cloud, local, or coding assistant — can read its contents and answer in one turn.
• "What's this?" beside an attachment is now understood as being about the attachment, not about whichever document happens to be open in the editor.
• Chats without a project folder no longer change files on your computer. The assistant can still read the files you attach, but if you ask it to create or edit files it will ask you to open a project folder first.
• Very large images are resized automatically so they can always be sent, and attachment copies made for the models are tidied up automatically after a week.

1.2.0 — 2026-08-02

• You can now speak your messages instead of typing them. Hold the spacebar in the chat box, say what you want, and let go — the words appear where your cursor was. There is a microphone button next to Send, and Cmd/Ctrl+Shift+Space turns dictation on hands-free so you can talk without holding a key. Nothing is ever sent automatically: you read what was heard, fix anything it got wrong, and send it yourself.
• On a Mac, your voice never leaves your computer. Speech is turned into text on the machine itself, so it works with no connection and nothing is uploaded.
• On Windows and Linux, dictation opens a small listening window in Chrome or Edge, which closes when you stop. One of those browsers needs to be installed.
• You can see that it is hearing you. A small level bar appears where your next word will land and moves with your voice, so a silent microphone is obvious immediately rather than after you have said a whole sentence.
• Say "scratch that" to undo the last thing you said, "new line", "full stop", or "stop dictation" to finish. A command only counts when you say it on its own, so ordinary sentences are unaffected.
• Dictation switches itself off after a pause, or after three minutes, so a microphone is never left listening by accident.
• Choose which engine to use under Settings → General, where more than one is available.

1.1.2 — 2026-08-01

• Coding agents can now see your document when you use the app on its own. Claude Code and the other coding-agent tools reach your open document through a small connection inside Circuitry. On a machine running only the desktop app — with no separate Circuitry Server installed — that connection refused every request, so an agent could talk to you but could not read the workflow, design or notepad you were asking about. It now works the same way it does with Circuitry Server installed. A tool you start yourself from a terminal asks your permission first, as it always has; one started by your own chat message does not ask twice.
• Agents read the document you are actually looking at. With more than one Circuitry window open, an agent could be pointed at the wrong one and report an empty document. It now follows the window whose chat you are using.
• Fixed the app hanging on quit. Closing Circuitry could leave it sitting there without finishing, if a coding agent's connection was still open.

1.1.1 — 2026-07-30

• The app now updates itself on Windows and Linux, as well as macOS. Previously a new release had to be downloaded and installed by hand on every platform. Circuitry now checks on startup, downloads in the background, and installs when you quit.
• Coding-agent tools now work on Windows. Claude Code, Gemini CLI and Codex CLI installed through npm were being found but could not be started, so the model appeared in the list and then failed the moment you sent a message.
• Fixed a coding-agent chat that could not send its first message. Chats set to start a fresh session per message — rather than keeping one open — failed immediately with Claude Code. Affected every platform.

1.1.0 — 2026-07-29

• Out of beta. Circuitry is now a 1.x release. Nothing you have built changes; the version number simply reflects that the core is stable.
• Code steps in five compiled languages. Go, Rust, C, C++ and Zig, alongside JavaScript and Python. The desktop app builds them on your own machine — it installs the compiler each one needs, on request, into its own folder, with no administrator rights and nothing else on your system touched. Every download is checked against the publisher's own fingerprint first.
• Once built, a step runs anywhere — including in a deployed webhook, with no compiler needed at the far end, because the build travels with your workflow.

1.0.0-beta.5 — 2026-07-28

• Double-click a file and it opens in Circuitry. The desktop app now handles the file types it can edit — Circuitry documents (.circuitry, .canvas, .notepad, .sheet), Markdown, text and source code, configuration files, web files, CSV/TSV and Excel spreadsheets, PDFs, images, Word documents and video — so opening one from Finder, File Explorer or your file manager brings it up as a tab. Edit it and save, and it is written straight back to where it came from. Opening files this way works whether the app is already running or not, and selecting several at once opens them all.
• Spreadsheets opened from a file save back to that file. A CSV, TSV or Excel file opened from disk now saves over itself, keeping your column names and storing the formulas you typed rather than their last calculated values. Previously saving one asked where to put it and never touched the original. Saving a multi-tab workbook to CSV or TSV tells you how many tabs could not fit.

1.0.0-beta.4 — 2026-07-03

• Coding-agent chats in the desktop app now match the web experience. Chatting with Claude Code shows a live step-by-step transcript (thinking, tools, results), streams replies word by word, honors the effort and model selectors, starts faster (the session is prepared while you type), and the Stop button reliably interrupts the agent. Previously the desktop app returned the whole reply in one block with no progress shown.

1.0.0-beta.3 — 2026-06-23

• The close button always closes the app now. The red window button (and Quit) reliably closes Circuitry, including when you have unsaved changes — choosing “Don’t Save” now quits as expected — and when you’re still on the sign-in screen. Previously the window could refuse to close in those cases.

1.0.0-beta.1 — 2026-06-04

• First public beta of the Circuitry desktop app for Windows, macOS, and Linux (Intel & ARM).
• Circuitry now tells you when a newer app, EServer, or CLI is available, and warns you (or stops a run) if your EServer is too old to work with the current editor.

EServer (tray / CLI / Docker)

1.4.0 — 2026-08-07

• Stopping or quitting tells Circuitry immediately. Connected windows see the server go Offline within seconds, instead of discovering it up to a minute later.
• Stop, then Start, works reliably. The server now fully releases its network port when it stops, so starting it again no longer fails with the port still in use.
• The server remembers your open project across restarts. This is the foundation of project handover: any of your devices can ask a server which project lives on it — including a cloud server that has just woken up.
• It knows when it is really being used. The server tracks genuine activity — terminal keystrokes and output, file operations, code runs — so a Circuitry cloud server can stay awake while a long task runs and sleep only when truly idle. Routine status checks don't count, so an idle machine still sleeps on schedule.
• Updating is recommended but not required — previous versions keep working without the faster offline detection and project memory.

1.3.0 — 2026-08-05

• Stopping is now something you say, not something guessed from a dropped connection. The server could not tell "the user pressed Stop" apart from "that window reloaded", so a reload killed whatever the assistant was doing mid-task. Stop is now sent explicitly, and a task whose window merely went away runs to completion and is waiting when you come back.
• A message sent mid-task reaches the assistant while it works, rather than being held until the task ends.
• One conversation across the chat and the terminal. The server now hands the same assistant session between the two surfaces instead of running one behind each.
• Work in progress survives Circuitry restarting. The server now keeps a short recording of whatever an assistant is doing, so a Circuitry window that reloads mid-task can pick the task back up — including everything that happened while it was away. Previously that work continued but was invisible, and the chat was left showing an answer that never finished.
• A task that fails now says why. If an assistant's process stops unexpectedly, the reason is held long enough for the chat to report it rather than simply freezing.
• Terminal output is kept for a client that comes back. The server now holds recent terminal output per session, so a Circuitry window that restarts reconnects and catches up instead of showing an empty terminal for a task that is still running. Updating is recommended but not required — the previous version keeps working, but the chat improvements in this release are the ones it cannot do, so a reload will still end a task and a message sent mid-task will still wait until the end.

1.2.1 — 2026-08-04

• Fixed chat attachments failing without a project folder. Attaching an image to a chat backed by a coding assistant no longer errors when no project folder is open — attachments are handed over privately on your own machine, and the copies are cleaned up automatically after a week.
• Chats without a project folder no longer change files on your computer. The assistant can read what you attach, but file creation and editing now ask you to open a project folder first. Updating is recommended but not required — the previous version keeps working.

1.2.0 — 2026-08-02

• No changes for Circuitry Server in this release. Its version moves with the desktop app so the two stay easy to talk about; there is nothing you need to do, and 1.1.2 keeps working.

1.1.2 — 2026-08-01

• Fixed coding agents reporting a permission problem that was really an expired key. The connection a coding agent uses to read your document signs in with an access key of its own. If that key had been reset — which happens when Circuitry Server re-encrypts its keys — every request failed, and the agent was told its connection had not been approved. It would then ask again, and again, each attempt costing you a full reply. Chats now hand the agent a key that cannot go stale, and nothing has to be set up by hand.
• Agents read the document you are actually looking at. With more than one Circuitry window or tab open, an agent could be pointed at a different one and answer as though your document were empty. It now follows the window whose chat you are using.

1.1.1 — 2026-07-30

• Coding-agent tools now work on Windows. Claude Code, Gemini CLI and Codex CLI installed through npm were being found but could not actually be started, so the model appeared in the list and then failed the moment you sent a message. They now launch correctly, whichever way the tool was installed.
• A tool that cannot run is no longer offered. On Windows, anything Circuitry Server cannot actually start is reported as not installed, rather than appearing as an available model that fails on first use.
• Fixed a coding-agent chat that could not send its first message. Chats set to start a fresh session for each message — rather than keeping one open — failed immediately with Claude Code. Affected every platform.

1.1.0 — 2026-07-29

• Four more languages for code steps. Alongside Go you can now write a step in Rust, C, C++ or Zig, and all of them run the same way — on your own computer, or in the cloud using the build saved with your workflow. Rust and Zig steps build to very small programs (tens of kilobytes), so they start instantly.
• Reading JSON is built in for every one of them. Rust, C and C++ have no JSON in their standard libraries and no package manager here, so Circuitry provides the type for you — you do not add a dependency, and you should not try to. Zig uses its own standard-library JSON.
• Circuitry installs the compiler each language needs, on request, into its own folder — no administrator rights, and nothing else on your system is touched. Every download is checked against the publisher's own fingerprint before it is used. C and C++ share one install.
• Your code is now a file of its own. Previously Circuitry placed your snippet underneath its own supporting code, which quietly meant you could only use the handful of packages Circuitry happened to bring in, and you could not add an import of your own at all. Now you write ordinary Go or Rust: import anything from the language's standard library exactly as you would in any other project.
• Compile errors point at your line. They used to be offset by the length of Circuitry's supporting code, so the marker landed past the end of what you had written.
• If you have existing Go steps, add the imports they use — for example import "fmt" if the step prints. The editor marks the exact line. Nothing else about them changes.
• Building needs a computer of your own connected — the desktop app, or a machine running Circuitry Server. Building in the cloud is not available yet, so a tablet on its own cannot build a compiled step; once built, it runs anywhere, including deployed webhooks.
• Building needs this version. An older Circuitry Server cannot build steps written the new way, so Circuitry asks you to update rather than showing a compiler error you did not cause. Everything else about an older server keeps working.

1.0.0-beta.5 — 2026-07-09

• Connecting a tablet or phone by QR code works again. Scanning the QR code from Circuitry Server failed with “Unauthorized”, because pairing asked the device for the access key that pairing itself is supposed to hand out. Scan the code and the device pairs; the QR window now confirms which device connected and tells you it's safe to close.
• Circuitry Server in Docker is reachable on the port you publish. The container ignored its configuration file and only listened inside itself, so every connection from outside was refused — while the container still reported itself healthy.
• Your access keys are better protected. They're now encrypted with a secret unique to your install rather than one shared by every copy of Circuitry Server, and they can no longer be read back over the network without authenticating first.
• You now pair a browser deliberately. Because of the change above, a browser that finds Circuitry Server on your machine no longer configures itself with a key it fetched from the server. Connect by scanning the QR code, or paste an access key under Settings → Developer → Execution Server. Devices you have already paired keep working.
• A server left running for weeks now hears about new versions. Circuitry Server used to check for updates only when it started, so one that had been running a long time could sit on an old build indefinitely. Circuitry now tells it when a newer version is published, and the tray shows a notification once — not a dialog, and not again for the same version. Older servers are unaffected and keep checking at startup as before.

1.0.0-beta.4 — 2026-07-03

• Coding-agent chats are faster and far more transparent. When you chat with Claude Code through Circuitry, the chat now shows a live step-by-step transcript (what the agent is thinking, which tools it runs, and their results), streams its replies word by word, and starts noticeably faster — the agent session is prepared while you're still typing. Long quiet spells now say why ("waiting on the model") instead of looking stuck, and Stop always stops. Update your Circuitry Server to get these in the web and iPad editors; older servers keep working with the previous chat behavior.
• Per-chat effort and backing-model controls now reach the agent. The effort selector under the chat input controls how deeply the agent thinks (off = fastest), and the model selector picks which Claude model the session runs on.

1.0.0-beta.2 — 2026-06-06

• Webhook listen mode. When you arm “Listen for request” on a webhook in the editor, your self-hosted EServer now captures the next incoming request and shows you its payload — without running the workflow — so you can see the exact shape of the data before wiring up the rest. Update your EServer to use Listen on self-hosted webhooks; older EServers keep working for everything else.

1.0.0-beta.1 — 2026-06-04

• First public beta of the Circuitry EServer.
• The tray app now has a Check for Updates… option and notifies you on startup when a newer version is available.
• The command line (eserver update) now reports versions correctly and shows an update notice when you start the server (turn it off with CIRCUITRY_ESERVER_NO_UPDATE_CHECK=1).

MCP server (@circuitry/mcp-server)

2.1.0 — 2026-07-02

• Picks up new tools automatically. The MCP server now fetches its tool catalog from your running Circuitry when it connects, so new Circuitry features appear in Claude Code and other coding agents without updating the MCP server. If a future Circuitry ever does need a newer server, it will tell you at connect time.
• Canvas screenshots arrive as images. Any tool that captures your canvas or screens now delivers a real image to the coding agent (not raw data), so agents that understand images can see your sketches and designs.
• New document tools. Coding agents can now inspect the active document (doc.describe, doc.query, doc.selection), see the canvas (doc.screenshot), draw and rearrange content (doc.patch — shapes, text, SVG art, align/distribute, grouping), and pull how-to guides on demand (skills.list / skills.get).

Web editor

2026-08-01

• Coding-agent replies start sooner, and stay quick as a conversation grows. Every message to a coding agent was preparing a large block of background information that the agent never receives, and re-sending the whole conversation to a tool that already remembers it — so a chat got steadily heavier the longer you talked. Both are gone: the material sent with each message now stays the same size from your first question to your fiftieth.
• No more pause before a coding agent starts. On a workflow, each message first asked a separate cloud model to work out which part of the canvas you meant — several seconds of waiting before your own agent was even given the message. Coding agents work this out themselves, so the wait has been removed. Chats using a cloud model are unchanged.

2026-07-26

• Turn text into editable letter shapes. Edit → Convert Text to Paths replaces a text object with a group of shapes, one per letter, with the counters — the hole in an 'o', the two in a 'B' — kept as real holes. Ungroup it and you can reshape a single serif point by point, colour half a word differently, or hand artwork to someone who doesn't have your font. Colour, opacity, effects, position, size, stretch and rotation all carry over, so the shapes land exactly where the letters were. It's one-way — undo restores the text — and text flowed along a curve has to be detached from the curve first.
• Text now takes part in combining and masking, with nothing to prepare. Select a word and a shape and choose any boolean operation: the letters are outlined for you behind the scenes. The Combine button used to refuse a selection containing text.
• Fixed: "Convert to Outlines" did nothing. The right-click item on a text object reported no error and made no change. It now works, and it's the lighter-touch alternative to Convert to Paths — the words stay fully editable and their outlines are recorded alongside, which is what lets an SVG export match your screen on a machine without your font installed.
• Clearer icons for the boolean operations. Unite, Subtract Front, Subtract Back, Intersect, Exclude Overlap and Divide now show the two-overlapping-squares picture used across design tools, with the part you'll keep filled in — so you can tell them apart without reading the labels. Help pages show the same icons.
• Stroke options — dashes, ends and corners. Select a shape, a line, or a path and the Style panel now has a Line Style section: dashed, dotted, dash-dot or long-dash patterns with a spacing control; flat, round or square line ends; and sharp, round or bevelled corners (with a sharpness limit for very tight angles). Dash lengths follow the line's thickness, so a pattern still reads as dashed whether the line is a hairline or a heavy band. Everything you set is kept when you export to SVG. Freehand pen strokes don't have these — they're drawn as ink, not as a line with a width.
• A help page for the Style panel. The panel that holds thickness, colour, fill, opacity, effects and now line style had never been documented. There's now a full page covering where to find it in each editor, what every row does, and how it behaves with several objects selected or with a group.
• Type exact sizes and positions. With something selected, Edit → Transform… opens a small panel with X, Y, width, height, and rotation. Type a value and the object changes as you type, so you can nudge a shape onto a precise size or line two things up to the pixel without hunting with a handle. The link button between width and height keeps the proportions, and rotation turns the selection by the number of degrees you enter, around its own centre.
• The panel stays out of your way. It opens beside whatever you selected rather than on top of it, and it doesn't dim, blur, or block the canvas — you can watch the object change while you type. Drag it by its title bar if it's still in the way. Available on the whiteboard, in notepads, in workflows, and in the Designer.
• Fixed: undo didn't undo. Resizing an object, clipping an image's transparent edges, or combining shapes could leave you unable to undo the change — and a single Transform edit could need several presses. One edit is now one press of Undo, everywhere.

2026-07-25

• Combine shapes (boolean operations). Select two or more objects and combine them into a single new shape: Unite, Subtract Front, Subtract Back, Intersect, Exclude Overlap, and Divide. If you know these as boolean or pathfinder operations, that's what they are — they are labelled both ways. Unlike grouping, which keeps your objects and moves them together, combining replaces them with real new geometry you can carry on editing point by point, and holes are genuine holes: cut a circle out of a square and you get a ring that keeps its hole when you move it.
• Subtract works in both directions. Which shape survives is decided by what's in front of what, so if the result comes out the wrong way round you don't need to restack anything — Subtract Front keeps the shape underneath, Subtract Back keeps the one on top.
• Reach it three ways on the same selection: the Combine (Boolean) button on the toolbar above the canvas, Edit → Combine Shapes (Boolean) in the menu, or the right-click menu. Available on the whiteboard, in notepads, in workflows, and in the Designer.
• Icons and vector images are converted for you — stamp a star icon over a rectangle, choose Subtract Front, and you get a star-shaped bite without converting anything by hand first. Pen and brush strokes combine by the line you actually see. Photos and editable text are left alone. The combined shape keeps the front object's colours and stays where it was in the stack.
• Fixed: Copy, Copy as Image, Save as Image, and Paste did nothing in the Designer. All four silently ignored your selection. Clipping an image's transparent edges in the Designer also reported success while discarding the result.
• Fixed: opening the File or Edit menu could crash the editor on some documents when objects were selected.
• Fixed: menu items with a submenu were unresponsive in the desktop app and on iPad.
• Vector export. Your artwork can now leave Circuitry as SVG, so it opens in any other design tool at any size. File → Export Screens… in the Designer offers SVG alongside PNG, JPEG, WEBP, and HTML — for the current artboard, all of them, or a hand-picked set — and can save to your downloads, a connected Circuitry Server, or a folder on this computer. The Designer's toolbar also has a one-click vector button that opens straight to SVG.
• Export just a selection. Pick a few objects and press SVG in the toolbar above the canvas to download only those, cropped to their own edges — handy for handing a single logo or icon to someone. On the whiteboard and in notepads, File → Export Selection as SVG… does the same thing; these editors could not export vectors at all before.
• What comes across: shapes stay real shapes, pen and brush strokes keep the pressure and taper you drew, gradients and shadows carry over, pictures are embedded so the file stands alone, and text stays editable text (convert type to outlines first if you need it to look identical on a machine without the font). Objects with 3D depth export as their flat outline, since vector files have no notion of depth.
• Fixed: the Designer's "Export as SVG" button did nothing. It was a placeholder that silently no-opped. It now works.
• Sharper drawings in exported web pages. When a design is exported as HTML, the drawn artwork embedded in the page is now rendered properly — previously most objects came out as thin hairline outlines that lost their shape, fill, and thickness.
• Help updated. Every editor tour now documents the toolbar that appears when you select something on the canvas, which the tours had never covered, plus the new export routes.

2026-07-18

• Stronger sign-up protection. Creating a free account now includes a quick automatic human-verification check (invisible for almost everyone), and the sign-up form no longer accepts throwaway variations of an email address that already has an account. This stops the automated fake sign-ups that were sending unwanted verification emails to real people's addresses.

2026-07-16

• Editor tours in the help. A new "Editors" section in the docs gives every editor a guided tour — Whiteboard, Workflow, Designer, Code IDE, Notepad, Chat, CodeBook, Spreadsheet, Web pages, 3D scenes, Terminal, and Browser — each showing the real screen, every control with the exact icon it wears in the app (including the pen tray's actual implements), verified keyboard shortcuts, and what each editor is best at. A companion "Selection Toolbar" page explains the button row that appears above the canvas whenever something is selected. The Workflow tour walks five real scenes: a batch image pipeline run, Python nodes with breakpoint debugging, input-aware node wizards, the Plugins Library, and AI-drawn flowcharts.
• Help fixes. The keyboard-shortcuts reference was rewritten to match the app exactly (temporary pan is Hold Shift — Space opens the node picker; many missing shortcuts added), the iPad gestures page no longer lists gestures that don't exist, and the getting-started guides now agree on how nodes connect (from any edge, with automatic routing).
• Keyboard navigation for tabs and split-screen panes. Cycle editor tabs with Cmd+Option+Right / Left (Mac) / Ctrl+Alt+Right / Left (Windows / Linux) and move focus between split-screen panes with Cmd+Shift+Option+Arrow (Mac) / Ctrl+Shift+Alt+Arrow (Windows / Linux) — both also in the View menu. Together with the existing arrow-key node navigation (arrows move between connected nodes, Enter edits, Esc steps out), the whole workspace is now reachable without the mouse.
• Menu keyboard shortcuts work again. Shortcuts shown next to View-menu items (Full Screen and others) quietly did nothing when pressed; the menu click always worked. Fixed — pressing the shown combination now triggers the menu action.
• Vim mode. Turn on Vim keybindings for every code editor in the app — code files, workflow node editors (Code, Agent, Data Viz, Code File, Params, Database, Web), docked side panels, notebook cells, page-source editors, and 3D scene code. Enable it in Settings → Code Editing or toggle from anywhere with Cmd/Ctrl+Shift+V. A slim status bar in each editor shows the current mode and follows the editor's light/dark theme. In workflow nodes, Esc is mode-aware: it leaves insert/visual mode first, and only exits the editor when pressed again in normal mode. Off by default. See the Vim Mode guide.

2026-07-11

• Pinch-zoom on the workflow canvas is smooth again on iPad. A recent change to touch handling made pinch-zoom process far more updates than the screen could show, so zooming a busy canvas stuttered badly (while panning stayed smooth). Pinch now updates once per frame, restoring full speed — including on Designer, Notepad, and chat flows.
• Panning a large canvas no longer re-loads nodes that scroll off-screen. Code and image nodes used to reset when they left the view and came back; they now stay put for normal-sized canvases.

2026-06-11

• Resolve import placeholders automatically. Placeholder nodes left by an import can now be converted into working nodes: Circuitry matches them against its Integrations Gallery (one click swaps in the matching operation with settings pre-filled), and an AI pass rebuilds the rest using a model you pick — including local and CLI models. Use Open & Resolve right after an import, or right-click a placeholder → Resolve Placeholder…. Everything resolved is flagged for review before you run.
• Import your work from other tools. A new File → Import… dialog converts files exported elsewhere into native Circuitry documents: n8n workflows, Make (Integromat) blueprints, Node-RED flows, Langflow flows, and Flowise chatflows become workflows; Excalidraw scenes become notepad drawings; Jupyter notebooks (.ipynb) become CodeBooks; and SVG files or Figma files (via file link + personal access token) become Designer screens. Imports are best-effort: steps without a Circuitry equivalent become clearly marked placeholder nodes that keep their original settings and wiring, and an import report shows what mapped cleanly, what needs review, and which service connections to set up again. See the Import guide.

2026-06-08

• Scheduled (timer) workflows now run in the cloud. A workflow with a Schedule start node set to cloud hosting now fires automatically on its timer — no need for an always-on EServer. Set it up with a friendly picker (every few minutes, daily at a time, weekly on chosen days, monthly on a day) instead of writing cron by hand, pick your timezone, and see the next few run times before you deploy. There's also a Run on schedule now button to test the flow immediately. Advanced users can still enter a raw cron expression.
• Updated Terms of Service (v1.3). We've clarified that Circuitry and its outputs may not be used to train competing AI models or build a competing product, that automated scraping of the app isn't permitted, and that third-party names and logos shown on integrations belong to their owners and don't imply any partnership. You'll be asked to review and accept the updated terms next time you sign in.

2026-06-06

• Inspect incoming webhook requests. Webhook start nodes now have a Listen for request button (and right-click menu item). Arm it, trigger your webhook from your other app, and the node lights up and captures the request’s payload so you can see its structure and build the rest of your flow against real data. The workflow itself doesn’t run while you’re just listening.
• The webhook URL now matches where you’ve chosen to host it — pick an EServer and you’ll see that server’s address instead of the cloud URL.

1.0.0-beta.1 — 2026-06-04

• The editor now checks that a connected EServer is compatible: it shows a banner when an update is available and blocks running workflows on an EServer that's too old, with a clear prompt to update.