0.1.13

Interactive documents

Added

  • Documents can now be interactive. Ask for a comparison, an explainer, or a small interactive report and Koda creates it as a self-contained HTML document that opens on the Stage, joins the Library and search like any other document, and takes the same star, hold view, and live refresh markdown documents have. Its scripts run in a locked-down sandbox with no network access and no reach into your files.
  • Select a passage in a document and turn it into an interactive view. Koda creates a sibling HTML document from your selection, links it back into your writing, and opens it beside the source. A link to a local document now shows as a small card you can open in place, and the file keeps a plain relative link, so the document stays portable in any other editor.
  • A new Usage screen in Settings shows what your whole subscription did, counting Claude Code and Codex activity on this Mac inside and outside Koda, terminal sessions included. Pick a window from 24 hours up to 90 days and see what those tokens would have cost at full API rates, what cache reuse saved you, and the split across providers and models. Every dollar cites its rate, and a model with no citable rate shows tokens instead of an invented number.
  • The agent can now create documents properly from chat on both engines, with their title, date, and kind in place, and the document it makes opens on the Stage straight away.
  • Path-shaped code in an agent reply, such as src/app.ts:12, can now be clicked to open that file on the Stage.
  • You can ask Koda to put a document on your Documents shelf, or take it off, in the middle of a conversation.

Changed

  • Search is now one door. ⌘K, ⌘P, and ⌘⇧F all open the Library, typing also searches every file in the project by name or by a phrase from inside, and find and replace lives right on the file you are looking at.
  • Your Documents shelf is now kept with the project itself, so moving the project folder or restoring an earlier version keeps your starred documents, and a star follows its document through renames and moves.
  • An open document's Stage bar keeps the current view in one labelled menu, with Hold view, Reveal in Finder, Copy path, and Delete inside. Starring stays visible.
  • AI providers in Settings now focuses on sign-in, billing, and your plan's live limit windows. The plan gauges in the bottom bar carry their reset clock inline, and a new Full usage row opens the Usage screen.
  • The sidebar now has one clear hierarchy: Sessions, Starred documents, and History.
  • Switching the Codex ChatGPT account now immediately refreshes the live usage windows.

Fixed

  • A session could get permanently stuck refusing new messages after an approval you never answered outlived its turn. Approvals now clean themselves up, and a message Koda refuses to send shows a banner explaining why.
  • Signing in to Codex inside Koda now also signs in the codex terminal command, so the terminal copy of your login no longer quietly expires.
  • Koda no longer refuses to write a markdown file that belongs to the code, and files that only look like markdown, such as .mdx, open in the file editor so their syntax cannot be quietly rewritten.
View release on GitHub
0.1.12

The Library

Added

  • Press ⌘K for the Library, one place to create or find anything you and Koda have written. Search by title, words inside the page, type, or date. You can also ask a question across your documents and past chats, and Koda lists the sources it used.
  • Say "keep this as a document" and Koda writes the useful result into your Documents folder. The document remembers which chat it came from, and a star keeps it in the sidebar across chats.
  • Documents open ready to read and switch to Edit when you want to type. Changes save as you go, and selecting a passage shows how much text will travel with your message to Koda.
  • Agents can now place a document, a source location, or a diff on the right chat's Stage, and links in replies open your project's documents and files on Mac or phone.
  • Each finished turn keeps its own This turn file set, so the exact files and diffs stay reviewable after the work is saved.
  • Settings lets you choose how Koda writes session names and saved-version descriptions, from Apple Intelligence on your Mac to plain local text or your configured Claude or Codex account.
  • Koda now includes Deep Review for changes that need a closer look. It follows related code, tests, and history, and gives the change an evidence-backed score.

Changed

  • The Stage now keeps several things open as tabs. An app, document, diff, terminal, and agent roster can stay together, and a narrow window keeps both the Stage and the conversation readable.
  • Versions is now one timeline. Unsaved work sits at the top, side work branches from the point where it began, and the GitHub boundary and push action sit on the same line.
  • A turn's commands and file work fold into one timed work line, and work sent to several agents opens a roster where you can inspect or stop each one.
  • The sidebar is a quieter chat list. Files open through ⌘P and documents through the Library.
  • Model selection now opens inside the current AI provider, with a short explanation for each curated model.
  • Usage now shows the full API value of your work, what caching saved, recent model costs, and a daily chart.
  • A successful overnight memory tidy now saves its own clean changes as one version credited to Koda Dream, so overnight saves stay easy to recognize.
  • Background checks slow down when Koda is out of sight or your Mac is on battery, and the bundled Codex engine is updated to 0.147.0.

Fixed

  • Codex work is no longer interrupted after a fixed number of steps, chats sort by what actually needs you, stopping a turn stops its delegated agents first, and new chats keep their full provider, model, and reasoning choice.
  • Many smaller fixes cover document recovery, session names, archived chats, saved-image cleanup, and screen-reader labels. The full list lives in the project changelog.
View release on GitHub
0.1.10

Finished work gets a version

When an agent finishes a task in a project that uses Git, Koda now saves it as a local version you can return to. Continuing work stays together, new coding topics start clean, and Versions tells you what is ready to review. You decide when anything is pushed, merged, or discarded.

Added

  • In projects that use Git, agents now save each finished task as a local

version you can return to. Continuing work stays together, and a genuinely new coding topic starts from the project's main line. The Versions view labels each workspace as Ready, Needs cleanup, or Needs a check.

  • Koda can now split independent parts of a job across fresh agents in

parallel. Each one has its own visible task card, and you can stop one without ending the rest of the session.

  • A fresh pair of eyes is now part of finishing substantial work. Screens,

pages, and documents are opened and checked against the agreed standard before you see them. Code changes get a bug review, and finished features get an architecture review for duplicate or competing parts. These checks use more of your plan and can be turned off in Settings → General → Finishing work.

  • Koda can now tidy project memory overnight and leave a dated journal for the

morning. Anything that needs your judgment is flagged for you. The feature is off by default and can be turned on in Settings → Memory.

  • Changes to Koda's guardrails, recovery store, or app settings now always ask

for your approval, even in Auto-approve. The approval card explains why.

  • Messages about Claude or OpenAI outages now link to that service's status

page.

Changed

  • Codex now shows command output and plan progress while it works. When it

shortens a long conversation, Koda says so and refreshes the context meter. Plan-usage percentages also stay in order when account updates arrive at the same time.

  • Agents now use the right built-in guide for each job and spend less of your

usage allowance rereading general instructions.

  • Documents opens with folders closed and remembers which ones you open. Folders

that contain no documents stay in Files, and the agent sees how Documents is already organized before it writes something new.

  • Saying no to an action now covers the outcome. The agent asks what you would

prefer next.

  • Dictation on macOS 26 now uses Apple's newest on-device speech engine for

sharper punctuation, names, and technical words. Older macOS versions keep the previous on-device engine.

  • The bundled Claude engine is updated to 2.1.221.

Fixed

  • Koda no longer replaces saved chats, archives, project lists, settings, or up

to 90 days of usage history when one of those files cannot be read. It preserves the original, keeps a recovery copy when needed, pauses risky writes, and tells you what happened.

  • Koda now tells you when a change was saved without an undo point. Edits that

would destroy existing work stop when the recovery point fails. A brand-new empty project also gets a proper first point in its recovery timeline.

  • Attaching an image now works the same from a drop or the attach menu. JPEG,

PNG, GIF, and WebP files that can be sent are available in both places. A file Koda cannot attach is named in the composer while the rest of the batch stays attached.

  • A downloaded update stays ready until you restart, even after another update

check or a stretch without internet.

  • Claude plan usage now shows the 5-hour, weekly, and per-model windows at

ordinary usage levels. A reset notification fires once when a limit truly clears.

  • A chat keeps its approval mode through reconnects, model changes, and other

background refreshes. Pending approval cards also return if Koda's window reloads.

  • Various fixes improve app relaunches, model selection, cloud sign-in recovery,

outage messages, document mentions with spaces, and guidance when you ask for a service Koda cannot reach.

View release on GitHub
0.1.9

Documents, done right

This release makes documents in Koda feel like real documents. Wide tables get room to spread out, long docs get an outline you can jump around, links actually work, and you can find text in a doc or save it as a PDF to share with anyone. Add in spellcheck, drag-out, and a few Mac niceties, and Koda fits your desk a little better.

Added

  • Documents got room to breathe. Wide tables no longer get crushed into the narrow reading column — they spread out toward the edge of the pane so a comparison actually reads like one. A new control in the top corner of any doc switches the whole page to full width (and remembers your choice per document).
  • Save a document as a PDF. File → Export as PDF… turns the doc you're reading into a clean, print-ready PDF — pick where to save it and it opens right up, ready to share with anyone who doesn't have Koda.
  • Find inside a document. Press ⌘F while working in a doc and a small find bar searches just that document, with next and previous. ⌘F in the conversation still searches the conversation, like before.
  • An outline for long documents. A subtle set of dashes on the right edge of a doc marks its sections — hover to see the section names, click one to jump there. It highlights where you are as you scroll.
  • Links inside documents now work. A link to another document opens it, a link to a section jumps to it, and a web link opens in your browser — before, clicking a link in a doc did nothing.
  • Spellcheck now works the way you expect. Misspelled words in the message box or a document have always been underlined, but right-clicking one did nothing. Now you get spelling suggestions, an Add to Dictionary option, and the usual Cut, Copy and Paste menu in any text field.
  • Drag files out of Koda. A file or folder from the Files or Documents list can now be dragged straight into Finder, Mail, or a browser upload box, the same way dragging into Koda already worked.
  • Session names now tell each other apart. Working on the same thing across several sessions used to name them all identically. Once a session's first round of work finishes it renames itself after what was actually done, then stays put — and if a name would exactly repeat one already in use, the date is added to tell them apart. A name you typed yourself is never touched.
  • Right-click the Koda icon in the Dock to jump straight to a recent project or start a new one.

Fixed

  • Links to your own documents open in Koda. When Koda writes a document and links to it in the conversation, clicking that link now opens it right there in the Stage instead of kicking you out to your web browser.
  • Resizing table columns in a document works again. Hover the border between two columns and drag to set their widths; the widths stick when you reopen the document.
  • Open Recent now brings an already-open project back into view. Previously, a project window that was hidden or minimized could make the menu look like it did nothing.
  • Adding a picture from Recent images no longer creates another copy of it.
View release on GitHub
0.1.8

Claude Opus 5

Added

  • Claude Opus 5, Anthropic's newest top model, now works in Koda. It came out

yesterday; if your model is set to Opus you get it automatically. The bundled Claude engine was updated to support it.

  • Do more with a change than just save it. Each file in your Changes list now

offers its own actions: see what changed, open the file to read or edit it, reveal it in Finder, or undo that one change. Hover a file for the quick buttons, or right-click it for the full list.

  • Open an image and see the picture. PNGs, JPGs, GIFs, SVGs, WebP and more now

show right in the workspace instead of a "binary file" notice.

  • Your Claude usage meter now shows a real percentage, just like the Codex one.

Anthropic recently started sharing exact usage numbers, so instead of a vague green-or-amber dot you see "79%" and can plan your day around it.

Fixed

  • The usage dot now actually turns amber when you're closing in on your weekly

Claude limit. Anthropic renamed the warning signal and its weekly window, so Koda was showing a calm green dot (and hiding the weekly gauge in Settings) even at 79% used. The tooltip also explains the 5-hour line instead of silently dropping it when there's nothing to report, and the "limit has reset" notification now covers the weekly limit too, not just the 5-hour one.

  • When your AI account is signed out, Koda now says so plainly and gives you a

Sign in button that takes you straight there, instead of a cryptic "please run /login" message with no way to act on it. The provider dot at the bottom of the window also turns amber with a "sign in" prompt so you catch it before you even send a message. It only ever shows for the provider you're actually using, so a Codex-only user is never nagged to sign into Claude, or the other way around.

  • Opening a project now starts you in your Koda folder, where your projects

actually live, instead of whatever folder your Mac last had open.

  • "Point at files or folders" in the attach menu now lets you select several at

once (Cmd-click or Shift-click), instead of making you go back for each one.

View release on GitHub
0.1.7

Koda tells you what it changed

Koda now closes out each piece of work by telling you, in plain words, what it changed, including anything it touched that you did not ask for. If it takes a quiet detour, you catch it right away, while undo is still one step away.

Added

  • Koda now ends each piece of work by telling you what changed in plain words, including anything it touched that you did not ask for. If it takes a quiet detour, you catch it right away, while undo is still one step away.
  • Pin the documents you use most. Right-click a document and choose Pin, and it stays at the top of your Docs list while keeping its place in your folders. Drag pinned documents to put them in the order you like.
  • See your documents by when they were last worked on. A new clock button above the Docs list flips it to a most-recent-first view, grouped by day, including documents Koda itself just wrote for you.

Changed

  • The new-project setup now gives you real room to describe your idea. The writing box is bigger, grows as you type, and the wording invites the full picture instead of a one-liner. The more you share up front, the better Koda sets up your project.
  • The light and dark switch is no longer in the bottom bar. You can still choose light, dark, or follow your Mac under Settings, then Appearance.

Fixed

  • The message box no longer collapses on you. In some cases it would shrink so far that your typing was pushed out of view, leaving only the top sliver of the placeholder text visible. It now keeps its proper height.
  • Your Codex usage meter now stays current. It could sit frozen at an old number, and on a plan with only a weekly limit it could show a mislabeled 5-hour window that never moved. Koda now checks your real usage after each turn and shows whichever limit actually applies to your plan.
  • Right-click menus in the sidebar now stay fully on screen. Opening one near the bottom or right edge of the window used to cut it off.
  • Codex keeps Koda's Preview and recovery tools available after the first message instead of silently losing them when it refreshes its tool connections.
  • Codex now opens pages and apps in Koda's Preview when it says it will. Its Preview instructions were naming a tool that did not exist, so it could promise to show you a finished page and then stop without opening anything.
View release on GitHub
0.1.6

Hand Koda your files

Koda can now work from your own documents and data, gives the pages it builds a real design, and helps you carry a long conversation into a fresh chat without losing the thread.

Added

  • You can now hand Koda files, not just pictures. Drag a CSV or PDF into the chat (or paste it), and Koda keeps a copy and reads it. This is great for importing data or working from a document. A new + button in the message box also lets you pick files, or point Koda at any file or folder on your Mac without copying anything.
  • Pages and screens Koda builds for you now go through a real design pass. It picks a look made for your project, with its own colors and type, instead of falling back on the same generic template every time.
  • When a conversation gets long, a "Keep going in a fresh chat" button now appears next to the fullness meter. Tap it and Koda writes a short handoff of where you left off, then opens a fresh chat carrying it over, so you can keep going at full speed without losing the thread. The old chat stays put.
  • The Mac File menu now has the project basics where you expect them. Make a document or folder, open a recent project, import files, reveal the project in Finder, copy its path, or close its window. Koda still saves continuously, so there is no Save command to remember.
  • You can now delete a project from the home screen. Hover over it in the recent list (or right-click it), type the project's name to confirm, and Koda stops anything the project is running and moves its folder to the Trash. You can always put it back if you change your mind.
  • Each engine's status now shows in the bottom bar. The small dot next to Claude or Codex is a health light. It stays calm when all is well and turns amber the moment that provider has trouble, with the honest word for it (outage, degraded, or maintenance). So when something feels off, you can tell at a glance whether it is you or them. It refreshes the moment you come back to the window.
  • The send button now does double duty, like you may know from other chat apps. When the message box is empty it is a microphone, so you can tap to dictate. As soon as you type or attach something, it smoothly turns into the send arrow.

Changed

  • The AI settings now give each provider its own tab. Pick Claude or Codex at the top and you see just that account and its usage, instead of one long stack of everything at once. Each provider's sign-in status shows right in its header, and the option to use your own API key is tucked behind an advanced toggle so the everyday view stays simple.
  • The light and dark switch moved from the strip at the very top of the window down to the bottom status bar, next to Settings. The top of the window stays calm, and day or night mode is still one click away.
  • When Koda asks you a multiple-choice question and you type your own reply instead of picking an option, the conversation now keeps showing the question you were asked, not just a note that you answered in your own words.
  • If you have Koda tell you when Claude is back after an outage, a single failed message is now enough to start watching. You no longer have to retry to make it notice. Anthropic's status page often lags behind what you are already seeing, so Koda watches quietly in the background from your first failed message and still catches the recovery once the page catches up.
  • Updated the bundled Claude engine to 2.1.205.

Fixed

  • The model you pick now actually runs. If you started a new chat and sent a message without re-opening the model selector, Koda could quietly run a different model than the one shown. You would pick Sonnet and get Opus. Your choice is now honored from the very first message.
  • Deleting a project now clears its chat history for good. Before, an old project's conversations could reappear if you later made a new project with the same name in the same place. Fresh start now means fresh.
  • If you delete a project's folder from your Mac, it no longer lingers as a dead tile in your recent projects. And if you bring the folder back, the project shows up again on its own.
View release on GitHub
0.1.5

A faster, tidier chat history

Koda now starts faster when you have a lot of archived chats, and you get to decide how long they stick around.

Added

  • Choose how long archived chats are kept. They stay forever by default. If you prefer to keep things tidy, you can now have Koda automatically delete archived chats older than a week, a month, or three months. The setting lives in Settings under Archived sessions.

Changed

  • Koda opens faster when you have a lot of archived chats. Archived conversations are now stored so that having hundreds of them no longer slows down startup.
  • The little status icons now tell you what is happening at a glance. When the assistant is thinking, its icon becomes a softly shimmering diamond that spins into a checkmark once it is done. When Koda is loading something, a small dot traces around the icon instead. Same quiet pixel style, easier to read.
  • The "memory needs a tidy" reminder now shows up far less often. Koda's assistant keeps its own notes about your project tidy as it works, clearing out finished items on its own instead of letting them pile up, so you are nudged to tidy only when it is really needed.

Fixed

  • When Claude asks you a multiple-choice question, your answer now stays on screen after you pick it. It used to show the choices again as if you never answered, so you couldn't tell what you'd picked. The card now keeps showing your selection.
  • The Codex usage details now label the 5-hour and weekly limits correctly. They could show up swapped, so the 5-hour limit appeared to reset days away. The reset times now line up with the right window, and when Codex does not say how long a window lasts, Koda no longer guesses "5-hour" for one that resets days from now.
View release on GitHub
0.1.4

Documents that behave like a Mac file list

Changed

  • Organizing Documents now works like a proper Mac file list. Folders can be

renamed, duplicated, revealed in Finder, or deleted from their visible menu; new documents and folders land inside the folder you selected, and new folders ask for a name right away. Arrow-key navigation and familiar rename, duplicate, delete, and new-document shortcuts work too.

  • Sub-folders in the Documents list now nest and indent under their parent, so a

folder inside a folder reads as a tidy tree instead of a full path written out on one line.

  • Selecting text in a document now opens one clear editing control instead of

separate formatting and agent-action bubbles. Format the text normally, or switch the same control into Koda's rewrite actions without losing your selection.

Fixed

  • Images in a document now show up in the document view. A picture you reference

by its file (like one in an assets folder next to the document) used to come up blank; now it renders inline where you wrote it.

  • Making a new folder in Documents no longer drops it inside a folder just

because you opened that folder to look inside. Use a folder's arrow to open and browse it, and click its name when you want new items to go there. The New folder and New document buttons now show where the new item will land.

  • The live Preview now opens when you use the Codex engine, not only the Claude

engine.

View release on GitHub
0.1.3

Point the agent at your documents with @

This update is a big round of everyday quality-of-life. You can now hand the agent a document just by typing @, bring files in straight from Finder, and right-click files and documents for the usual Mac shortcuts. Koda also keeps your project notes and Documents tidy on its own as things grow.

Added

  • Point the agent at one of your documents by typing @ in the message box.

Start typing the name, pick it from the list, and a reference drops in, so you no longer have to hunt the file down and open it first.

  • Drag files in from Finder to add them to your project. Drop them on a folder

to file them there, or anywhere in the panel to bring them in.

  • Right-click a file or document for the everyday Mac shortcuts: Reveal in

Finder, Copy path, Open in its usual app, and Duplicate.

  • Manage your writing straight from the Documents list too. Rename or delete a

document, drag one onto a folder to file it, and make new folders right there. A deleted document can be brought back from the recovery timeline.

  • Archived sessions (Settings → Archived sessions) now expand to preview the

last few messages, so you can recognize which chat is which before you restore it.

  • Koda now tells you when the agent's project notes have grown heavy enough to

slow it down. A "memory needs a tidy" notice appears in the status bar, and Settings → Memory shows the size and offers a one-button tidy that asks the agent to condense its own notes. Nothing gets deleted, and the tidy is undoable like any other change.

Changed

  • The agent now keeps its project notes lean as your project grows. It updates

the note it already has on a topic instead of piling on a new one, folds notes about replaced approaches into the current one, and keeps its running list of work to one line per item. Long-running projects stay fast instead of slowly filling up.

  • The agent now keeps your Documents organized as they add up. It groups related

documents into folders instead of leaving a pile of loose files at the top, puts each new document where it belongs, and offers to tidy up when the folder gets cluttered.

Fixed

  • Right-click menus (like Rename or Archive on a session) now have a clearly

defined edge in the light theme, instead of blending into the background.

View release on GitHub
0.1.2

Shape the idea before Koda builds it

When you ask for something new, Koda now helps you shape it first instead of charging at a guess. Plus a cleaner update banner and a couple of fixes.

Added

  • When you ask for a whole new app or feature, the agent now helps you shape it

first. It asks the few questions that actually change what gets built, suggests things you might not have thought of, and points out what a first version leaves out so you can ask for it next. It also will not talk you out of a big idea just because it sounds like a lot of work.

Changed

  • The update banner and its download progress are cleaner and easier to read.

Fixed

  • The model picker no longer jumps while you are reaching for it. The OpenAI

models used to pop in a moment after the menu opened and shove the Claude models up under your cursor. The menu now opens at its full height right away.

  • When a longer background task finishes while you are away, its result now

shows up on your next message instead of getting lost.

View release on GitHub
0.1.1

Maintenance update

A small maintenance release. Nothing changes in how you use Koda day to day.

  • Update delivery moved to the main Koda repository, so future updates come from the same place the app's source lives.

Requirements: an Apple Silicon Mac (M1 or later) on macOS 11 or newer. Signed and notarized.

View release on GitHub
0.1.0

Meet Koda

First public build of Koda, a guardrailed AI coding desktop app for macOS.

Koda runs on your own subscription: Claude Code with a Claude Pro or Max plan, or OpenAI Codex with a ChatGPT plan. You pick the engine and Koda handles the guardrails.

Requirements: an Apple Silicon Mac (M1 or later) on macOS 11 or newer. Signed and notarized.

What's in this release

  • In-app auto-update: Koda checks for new versions on launch and in the background, downloads them quietly, and asks you to restart when one is ready. Nothing installs on its own.
  • "What's New" appears once after an update with a short summary of what changed.
  • Settings shows the Koda version, the bundled Claude engine version, and a "Check for updates" button.
View release on GitHub