Every release. Every fix. On the record.

connect-v0.1.0

Clotho - A hands-on editor

• new**Reorder rungs by dragging** — grab a rung's number handle and drag it above or below another rung; a drop indicator shows exactly where it will land.
• new**Drag instructions straight from the palette onto a rung** — drop an instruction at a rung's start, end, or between existing elements, with a green indicator marking the insertion point. Drops are checked for basic ladder validity (inputs before outputs), so an illegal placement is caught with a clear message instead of producing broken logic.
• new**Edit tag descriptions inline** — change a tag's description right in the ladder view, no round-trip through a separate dialog.
• [improved] **Edit rung comments in place** — comment editing now happens in a proper inline box instead of a browser pop-up: Enter commits, Shift+Enter adds a line, Esc cancels, and clearing the text removes the comment.
• [improved] **Your edits save themselves** — ladder changes now auto-save a moment after you stop editing, so you never lose work to a forgotten save.
• fix**Long rungs wrap to fit the pane** — wide rungs now stack onto multiple rows that fit your screen instead of forcing horizontal scrolling, and output coils line up against the power rail so every rung shares a clean right edge, matching how Studio 5000 draws them. A wrap toggle lives in the ladder toolbar.
• fix**Timer and counter presets read correctly** — timers and counters whose Preset/Accumulator were stored as `?` now show their real tag values, exactly as the vendor IDE displays them, while your exported file stays byte-for-byte faithful.
• new**A live Problems panel** — a new panel lists issues grouped by routine, with error / warning / info filters, and marks the offending rungs and elements inline. Press Ctrl+Shift+V to verify the whole project and F4 to jump to the next problem. It checks your working copy — including unsaved tabs and edited-but-closed routines — not a stale reimport.
• [improved] **Smarter, quieter linting** — you're warned about rungs that drive no output, wrong operand counts, duplicate coils, undefined tags, unknown instructions, and operand-type mismatches — now with far fewer false alarms. Variable-length instructions like PID no longer trip the operand check, and I/O-module addresses are no longer mistaken for undefined tags.
• [improved] **The recognized Logix instruction set grew from 88 to 238 instructions** — timers and counters and their reset / with-input variants, FIFO/LIFO, Scale-with-Parameters, the full motion set, GuardLogix and drive-safety, ASCII/serial, and advanced math and trig are all understood now, so the assistant and the linter work correctly on real production programs.
• fix**No more bogus warnings on standard instructions** — SWPB's WORD/HIGH/LOW keyword operand, the alarm and immediate-output instructions (ALMD, ALMA, IOT), and the sequencer instructions (SQI, SQO, SQL) were being flagged as undefined tags or wrong operand counts on real projects; they now read correctly.
• new**Project-aware starter suggestions** — an empty chat now offers starting points drawn from your project's programs, routines, and tags, instead of generic prompts.
• new**Guided skills** — pick a task like Generate a Software Design Specification or Standardize comments across programs right from the chat box, and the assistant runs it with the right tools.
• new**Staged edits you approve before they land** — when the assistant proposes changes across several files, they gather in a review panel where you preview and then apply, instead of files changing out from under you.
• new**Reference files the assistant quotes exactly** — attach authoritative reference files to an assistant so it quotes them verbatim rather than paraphrasing from a fuzzy match.
• [improved] **Honest results** — when the assistant sets a value that already matched, it now tells you nothing changed instead of claiming it made an edit.
• [improved] **Smoother streaming** — answers replay as steady progress rather than an abrupt dump, auto-scroll respects it when you scroll up to read, and a long document that gets cut off is still shown with a clear "incomplete" marker instead of disappearing.
• [improved] **Tidier chat** — the assistant's background lookups collapse into a single group while edits and errors stay front and center, and the message box floats with a subtle fade and a spinner while an existing conversation loads.
• fix**"Ask AI about this tag" actually answers** — this now opens a real conversation and streams the answer into the chat, instead of quietly doing nothing.
• fix**Graceful handling of oversized requests** — when a project or request is too large for the model, you get a clear explanation instead of a confusing error, and there's no Retry button that would just fail the same way again.
• [improved] **Software Design Specifications now handle very large projects** — instead of failing or getting cut off when a project was too big to process in one pass, the generator now summarizes each program on its own, stitches the cross-program tag data-flow together, and assembles one coherent spec. You can watch it build live, and it's on by default.
• [improved] **Documents update in place** — asking to change an existing document now revises it rather than regenerating from scratch, and the tag reads and writes it reports are taken from the actual logic instead of being guessed.
• [improved] **The tool builder now lives in the main chat, with live preview** — your custom tool's interface renders as it's generated, and a "New chat" button lets you start fresh without losing the tool you're building.
• new**Custom tools can do far more** — they can now search and validate your project, trace logic, find where a tag is written, spot unused routines and dead rungs, pull text out of a PDF, and propose project edits that you review as a diff before anything is applied.
• [improved] **Tools open in a focused pop-up** — running a tool now opens a dedicated window rather than adding another workspace tab.
• new**A clear, usage-based credit system** — AI usage is metered by what it actually consumes, and a credit indicator in the status bar shows your remaining balance at a glance, with a breakdown of your cycle, add-on, and buffer credits and a link to billing.
• [improved] **Running low no longer stops you cold** — instead of a hard error, you get a gentle prompt; your action is held, and it resumes on its own once you top up. Optional auto-refill and a usage page with export are available, and on team plans a member can request a top-up for an admin to approve.
• [improved] **I/O Configuration as a proper tree** — your I/O now appears as a Studio-style hierarchy (backplane → networks → chassis → modules); click a module for a focused detail view.
• new**Reset a project to its original import** — rebuild a project from the exact file you imported, which is the way to re-parse an older import with the latest improvements.
• [improved] **Tooltips everywhere they help** — icon-only buttons across the app now carry clear tooltips for easier, more accessible navigation.
• [improved] **A more responsive layout** — sidebars toggle like a modern IDE with clearer active-tab indicators, and the resize handles are smoother with larger, easier-to-grab zones.
• fix**Layout fixes** — the chat input no longer shows the page background through it, and the project name no longer overlaps the toolbar buttons on narrow screens.
• new**Import Siemens TIA projects** — drop in a TIA Openness (SimaticML) export and open it as a project, then view and edit its blocks, data types, tags, and logic in the same editor you use for Rockwell. A first step toward first-class Siemens support alongside Rockwell.

Ariadne - Reads your logic

• newThe assistant can now read your **Function Block (FBD)** and **Sequential Function Chart (SFC)** routines — block-and-wire netlists (including feedback/latch loops) and SFC steps, transitions, and actions — so you can ask about graphical logic the same way you ask about ladder.
• [improved] FBD routines open in a rebuilt graphical viewer with pan, zoom, fit-to-view, and a dark theme — and your original block positions are kept exactly as drawn.
• [improved] FBD wires now route automatically as clean right-angle connections that go around blocks instead of through them, with tidy hop markers where wires cross.
• fixCross-reference and signal tracing now work through FBD routines and through Add-On Instruction blocks inside them, instead of stopping dead at the first function block.
• fixOpening older or hand-edited SFC routines that were missing transition or branch data no longer crashes the viewer.
• newYou can now **import Rockwell L5K files** — drop a `.l5k` export in alongside L5X, PLCopen XML, and PLCX and it opens as a full project.
• new**Edit and export L5K projects straight back to L5K.** Untouched parts of the file are preserved exactly; only what you changed gets regenerated.
• newAI chat and ladder editing now work on L5K projects, just like they do for L5X.
• newYou can attach a `.l5k` file as a chat context document, not just L5X.
• [improved] Projects now clearly show whether they came from an L5K or an L5X file — in the workspace pill, project switcher, project list, command palette, import badge, and export buttons.
• fixL5X files are no longer occasionally misread as L5K, and unusual lines in a Rockwell file (a malformed data-type member or tag) are now reported as a warning instead of being silently skipped on import.
• [improved] The assistant now works in **stages** — it plans your request, writes the change, then reviews its own work against PLC conventions before anything lands in your project.
• newWhen the reviewer spots a problem, the assistant **revises the change once on its own** and tells you in the reply (e.g. "Reviewer flagged 2 issues; auto-revised once").
• newPick an assistant **profile** right from the chat box — *Studio 5000* for the full pipeline, or *Studio 5000 Lite* for a leaner, faster pass.
• newCustomize the pipeline: turn each stage on or off, edit its instructions, and choose its model.
• [improved] **Ask and Plan modes are now strictly read-only** — the assistant can inspect your project but can never change a file in those modes, no matter how the request is phrased.
• fixAsking the assistant to read or quote a file (e.g. "read the README and quote its first line") now reliably reads the file instead of failing with an "invalid change" message.
• [improved] When the assistant shows you an existing routine, it now links to the file instead of pasting the whole thing back into chat — responses are faster and large routines no longer get cut off mid-answer.
• [improved] The assistant can work through longer, multi-step edits (read → analyze → modify → validate → commit) in a single request without cutting off partway.
• [improved] Your chat history now keeps the assistant's tool steps after a reload, so it remembers what it looked at earlier in the conversation.
• [improved] When the assistant says it will edit something but produces no changes, you get a clear notice with a **Retry** button instead of a silent dead end.
• [improved] Long blocks of code or output in chat now collapse into a tidy preview with a "Show all" toggle.
• [improved] The file you're editing is now always shared with the chat as a live context chip — no more accidentally dismissing it and getting answers about the wrong file.
• newThe assistant can now **edit the ladder logic inside your Add-On Instructions**, not just program routines — ask it to add, change, or remove rungs in an AOI and the edits apply right into the instruction.
• newNew **AOI Logic Diff viewer**: when an AOI changes, open it from the Changes view for a side-by-side comparison that shows only the routines that changed, lines up rungs by number, and color-codes what was added, removed, or modified.
• [improved] Scrolling through big ladder routines is now smooth and stable — no more flicker, jumping, or blank gaps on large files, and you can always reach the very last rung.
• [improved] Wide ladder rungs now scroll sideways instead of being squished or cut off, and they look the same in the editor, chat, diffs, and the Git preview.
• fixCopying or duplicating a rung now reproduces the original exactly — its comment and full tag paths (like `Track.Zn_AngRol[10].Transfer`) are preserved instead of being truncated.
• fixCopy a **range** of rungs even when the rung numbers aren't continuous, instead of getting told the rungs couldn't be found.
• fixCopy a rung into a **different routine** and it actually lands there — before, the copy looked like it applied but quietly disappeared.
• newDuplicate a rung **many times at once** — "duplicate rung 2 twenty times" now gives you twenty copies.
• newTell the assistant **where** a duplicated rung should land — e.g. "duplicate rung 1 and put it after rung 5."
• fixThe assistant now builds rungs with **parallel (OR) branches** correctly instead of flattening them into a single line.
• [improved] Exported **L5X files now re-open cleanly in Studio 5000** — module ports, connections, keying, tag data, alarm configs, and other controller settings survive a full round-trip through PLCcode, even on edited projects.
• fixUnsaved edits are no longer silently dropped on export when a routine was opened from the file tree instead of a chat link.
• fixExports no longer attach the wrong original data to nameless I/O modules after a module is deleted or reordered (modules are matched by catalog number, parent, and slot rather than position).
• [improved] Raised the upload limit for PLC project files to **200 MB**, so even very large projects import and export without hitting the size cap.
• newNew accounts now start with a ready-made **sample project**, so you can explore and learn the editor without importing a file first.
• newEach AI tool now has its own **Settings** — pick which model it uses and which project it focuses on, saved per tool.
• [improved] The **Process Flow Generator** was rebuilt: diagrams stream in and render live, you can Stop or Regenerate any flow, results are saved between sessions, and it uses far fewer credits.
• fixThe **IO List Extractor** no longer fails on large I/O trees — analysis streams in batches with a Continue button, and logs anything it skipped.
• fixThe **spec, SDS, and commissioning-checklist generators** no longer report 0 programs and 0 routines — they now count your project correctly.
• [improved] Spec generation streams its output live as it writes, and you can pick exactly which programs to summarize.
• fixProcess-flow diagrams and the PDF extractor now actually render — they were being silently blocked from loading the libraries they need.
• fixTool outputs now save correctly — writes from a tool's Outputs tab were silently failing.
• [improved] Click a tool in the file explorer to run it right away; right-click for Run or Edit. Reopening a tool tab now refreshes its contents correctly.
• newThe assistant can now **trace logic across your whole project** to answer questions like "why won't this output turn on" or "what gates this" — following each rung condition back to the first real blocker, with the exact tag and location, including FBD sheets.
• fix**Tag scope** now displays correctly everywhere: the tag browser shows only the tags for the scope you clicked, "where used" works for program-scoped tags, and the ladder tag picker offers the project's real controller and program tags.
• fixCross-references now resolve correctly for subscripted and member tags like `InputData[2].0` or `Timer_1.DN`.
• [improved] Tag cross-references load automatically when you select a tag — no more clicking "Find where used" — and the full list is paginated so every reference is reachable.
• fixClicking a cross-reference scrolls straight to the right rung and highlights it, even in long routines where it used to land off-target.
• fixThe Tags view now remembers your scroll position and selected tag when you switch tabs and come back.
• [improved] PLCcode.ai has a **refreshed look** — a dark amber-on-charcoal theme with new typography across the editor, sign-in, and Git pages.
• [improved] Clearer sidebar labels: **Workspace is now "Code"** and the edge-device area is now **"Wire,"** with a refreshed preview of what's coming.
• [improved] AI model names in the chat and tool pickers now show clean, consistent labels, and retired or unavailable models no longer clutter the list.
• fix**Fixed a rare issue where your session could briefly show another user's account.** Your sign-in now always stays your own.
• fixEditor tabs that got stuck on a loading skeleton after a page reload now reload their content on their own instead of staying blank.
• fixOpening a project built from an unusual or partly malformed L5X file no longer crashes the editor — tags and data types with missing names are handled gracefully.
• [improved] Rockwell **ACD files are no longer offered** as an import format, since they were never actually supported — the importer now lists Rockwell L5X, PLCopen XML, and PLCX so it's clear what works.
• newRequesting beta access now includes a quick checkbox to agree to our **Privacy Policy and Terms of Service** before your request goes through.
• newWhen you sign in, you'll be asked once to review and agree to our latest Privacy Policy and Terms of Service — and again if we update them.
• [improved] The confirmation email after requesting access now links to the policies you agreed to, so you can review them anytime.

Confirmations & polish

• newYou now get an email confirmation when your report is filed
• newReport status updates flow back to you as things move along
• newProper social share previews, icons, and page metadata
• fixSmoother validation and feedback in the report dialog

Signs of life

• newA live activity strip on the homepage showing ongoing development
• newOpen and resolved counts for reports filed through the site
• newTypical time-to-fix is now shown alongside the report counts
• perfReworked the header actions for a cleaner, quicker report flow
• fixTidied page spacing and layout across the homepage

A public feed of what we ship

• newThis updates page — a running log of what's changed
• newEach entry is curated before it goes public, so only the user-facing story shows
• newVersions are dated and labelled automatically as we publish
• perfUpdates are cached and refresh in the background, so the page stays fast

Report a problem, right from the page

• newReport a bug or request a feature without leaving the site
• newWrite-ups support Markdown with a live composer and preview
• newAdded a privacy page explaining what we collect and what we never store
• newYour contact details stay private and are never shown publicly