Repository navigation
v0.14.2
v0.14.2
Mac: capture goes back to SCScreenshotManager — and v0.14.1's explanation was wrong.
v0.14.1 said XDR screenshots failed because a 1600 downscale came out an odd 1039, and ScreenCaptureKit
refuses uneven dimensions. That was not the cause. A 1280×827 shot — an odd height — has since been
shown to work on the development Mac, so the even rounding goes out with the theory. The real cause inside
SCStream was never pinned down, and this release stops guessing at it.
These binaries carry no credentials. No CodeFlying key, no voice-provider key, and no Google key.
You supply your own: sign up from the link inside the app, then paste the key into
Settings → Backend. Starting is free and the free allowance is finite.Each package already points at the right server for its language — the English ones at
myapp.codeflying.app, the Chinese atbaas.kuafuai.net. You do not choose this.The macOS packages are signed with a local identity, not an Apple Developer ID, so Gatekeeper
refuses the first launch. On macOS 15 the old "right-click → Open" trick no longer exists; see
Opening the Mac app.
Correction to v0.14.1
v0.14.1's diagnosis was wrong, and the changelog and release notes for it said so in detail. This is the
correction.
| v0.14.1 said | Actually |
|---|---|
| A 1600 downscale produced an odd height (1039) | True — but not the cause |
| ScreenCaptureKit refuses uneven dimensions, so every shot failed | The even rounding it added fixed nothing |
| → therefore, force the dimensions even | The even rounding is removed |
What is known instead: a 1280×827 capture (odd height) worked on the development Mac. So an odd
dimension is not what broke it. SCStream failed only on downscaled captures, only on one machine,
and why was never found — so this release puts back the API that worked rather than keeping a
workaround for a cause nobody identified.
If you read v0.14.1's notes and acted on them, nothing is broken by this — the even rounding only ever
cost a pixel. It just was not the fix.
Highlights
- Capture is the one-shot API again, which is the path that worked before 2026-11-05.
- The build refuses to make a second capture path. An SDK older than 14 now stops the build instead of
silently producing a Mac app that takes screenshots a different way. - The two things v0.14.1 got right are kept: a capture failure says it is not a permission problem, and
the real error is written to a log file. That log is what made this correction possible.
What's Changed
Fixed
-
Capture is
SCScreenshotManager.captureImageagain, not a one-frameSCStream. The stream was a
stand-in introduced on 2026-11-05 so the bridge would build on a 13 SDK —SCScreenshotManager
is macOS 14-only. That date is exactly when the XDR failures began:- every downscaled shot failed while native-size ones still worked, on one specific machine;
- and the Kotlin side reported each failure as a missing Screen Recording grant, sending the user to
System Settings for a permission that was already valid.
Going back to the one-shot API restores the path that worked.
-
The even rounding is removed.
max(2, Int(w) & ~1)was added in v0.14.1 for the odd-height theory,
and that theory does not hold — so the code that was justified by it goes with it.
Changed
buildBridgenow refuses an SDK older than 14 instead of building a capture path of its own. The
rule it enforces: two packages of the same version must not take screenshots differently. That is
precisely how a build-time convenience became a user-visible bug — the workaround was invisible in the
source and wrong on one machine. The check stops the build with the fix in the message
(softwareupdate --list, or reinstall Command Line Tools).- This means the Mac app needs the macOS 14 SDK (Command Line Tools 15+) to build. It still needs
no Xcode — only the SDK, becauseldalso comes from Xcode. This machine has SDK 14.4, so the
release build passes.
- This means the Mac app needs the macOS 14 SDK (Command Line Tools 15+) to build. It still needs
Kept from v0.14.1
- A native capture failure says it is not a permission problem.
CAPTURE_FAILED_NOTE, and every caller
has already checkedscreenCaptureAllowedbefore a capture can fail. - The real
NSErroris appended to~/Library/Application Support/Andee/screenshot_error.log. A
GUI-launched app's stderr is/dev/null, so without this the error was lost entirely — and without it,
this release could not have been diagnosed.
Documentation
- The Mac build section of both READMEs,
CLAUDE.mdanddesktop/native/README.mdnow state the SDK
requirement and the whole history: the one-frame stream, why it was introduced, what it broke, and the
rule — "do not bring a second capture path back behind an SDK check."
Known issues
- 🔴 The fix has not been run on the machine that had the problem. The failing machine is a 3024×1964
XDR MacBook; this build was made on a different one. What is verified here is that the code is the API
that worked before the regression, that the build passes the new SDK check, and that the new Swift code
is really in the packaged app (thelibandeemac.dylibinside the jar differs from v0.14.1's and
contains the new strings). Whether the XDR machine now captures is the thing to check. - The root cause is still open.
SCStream's downscale failure on that machine was never explained.
This release routes around it. If it ever reappears, the log file above is where to start, and the
question to ask is now narrower: does the one-shot API fail the same way? - A Mac build now needs the macOS 14 SDK (Command Line Tools 15+). On an older toolchain the build stops
with a message naming the fix, by design — but that is a real new requirement for anyone building the Mac
app from source. - Only the Mac is affected. The Android packages are rebuilt so the version number stays in step;
nothing inapp/orcore/changed. - The task lifecycle has not been run end to end with a real model (v0.14.0), merge-undo has not been
exercised over its 30 days, and Android 8.1–10 still cannot see the screen (v0.13.0) — that path
still has not been run on a real 8.1–10 device. - Neither pet has been lived with for a day, and the pet's picture library is Mac-only (v0.12).
- The ball's face has never been verified on real hardware here. Carried from v0.8.1: the eyes are 3D
capsules under perspective projection, so a flat rendering cannot stand in for the real thing. - What it keeps for itself is per device. The one on your phone and the one on your Mac each have their
own record. They do not share a notebook. This is deliberate rather than pending. - Nothing verifies that a key is accepted by the server it is sent to. These packages are checked for
the right server and the absence of credentials; whether a given CodeFlying key works against
baas.kuafuai.netversusmyapp.codeflying.apphas not been exercised end-to-end. - The Mac still falls back silently when no server is configured. Every package here carries one, so it
does not arise for these downloads — but a locally built app with an empty domain resolves to the local
backend and sends whatever key it has toapi.deepseek.com, while the self-check reports "Configured". assets/andee_key.txtis tracked by git and is not in.gitignore. It ships with a single newline,
which reads as empty. A key written into it would outrank every build setting.- Two issues from
v0.1.0are unchanged and untouched:tools/verify_faces_lab.pystill reports
7/16 assertions passing, and the debug WebSocket server (net/BodyWsServer) still binds0.0.0.0:9008
with no authentication (documented inSECURITY.md).
Before you install
Read DISCLAIMER.md and
docs/acceptable-use.md first. On Android the app asks for accessibility,
overlay, notification access, SMS and call log, and it can complete a payment by tapping; on the Mac it
asks for accessibility, screen recording and notifications. The only guardrail on that path is Android's
FLAG_SECURE; the prompt-level checks on risky actions are instructions to the model, not enforcement.
This release changes how the Mac takes a screenshot, not what it may do with one. It still needs
Screen Recording; the difference is that it goes back to the API that worked.
Download
English packages
Android (Android 8.1 / API 27 or later). Take the one that matches your device's processor:
| Label shown on the page | File name | Take it if |
|---|---|---|
| Andee 0.14.2 · Android-arm64-v8a | andee-0.14.2-en-android-arm64-v8a.apk |
any phone from the last several years — almost everyone |
| Andee 0.14.2 · Android-armeabi-v7a | andee-0.14.2-en-android-armeabi-v7a.apk |
an older 32-bit device |
| Andee 0.14.2 · Android-x86_64 | andee-0.14.2-en-android-x86_64.apk |
an emulator on a computer |
Not sure? Take arm64-v8a. If it refuses to install because no architecture matches, try
armeabi-v7a — the two do not conflict, and only one of them can be right.
macOS (macOS 14 or later). Not a universal binary — take the one matching your Mac's chip:
| Label shown on the page | File name | Take it if |
|---|---|---|
| Andee 0.14.2 · Mac-Apple silicon | andee-0.14.2-en-mac-arm64.dmg |
an Apple silicon Mac (M1 or later) — most Macs |
| Andee 0.14.2 · Mac-Intel | andee-0.14.2-en-mac-x64.dmg |
an Intel Mac |
To check: Apple menu → About This Mac. "Chip: Apple M1/M2/M3/M4" means Apple silicon; "Processor:
Intel Core" means Intel.
Chinese packages
The Chinese packages are the -zh- files, and each displays a Chinese label on the Release page:
| Label shown on the page | File name |
|---|---|
| Andee 0.14.2 中文版 · Android-arm64-v8a | andee-0.14.2-zh-android-arm64-v8a.apk |
| Andee 0.14.2 中文版 · Android-armeabi-v7a | andee-0.14.2-zh-android-armeabi-v7a.apk |
| Andee 0.14.2 中文版 · Android-x86_64 | andee-0.14.2-zh-android-x86_64.apk |
| Andee 0.14.2 中文版 · Mac-Apple 芯片 | andee-0.14.2-zh-mac-arm64.dmg |
| Andee 0.14.2 中文版 · Mac-Intel 芯片 | andee-0.14.2-zh-mac-x64.dmg |
Which one to take is the same question as above. The only other difference is the server, which is
already baked in.
The file names are ASCII because GitHub deletes Han characters from release asset names, so the Chinese
wording lives in the label where it survives — and the downloaded file name stays ASCII, which means it
survives every filesystem it is copied onto as well.
Opening the Mac app
The .dmgs are signed with a local identity, not an Apple Developer ID, so Gatekeeper refuses the first
launch of a downloaded copy.
override, and most advice online still recommends it. The current route:
- Double-click the app. It will be refused. Dismiss the dialog.
- Open System Settings → Privacy & Security and scroll to Security.
- There is a line naming Andee as blocked, with an Open Anyway button beside it.
- Confirm, and authenticate.
That button only appears for about an hour after the failed launch. If it is missing, launch the app
again and go straight back to Settings.
One more thing worth knowing: the app inside the mounted dmg carries no quarantine flag, so it may
launch straight from the dmg — and then be blocked once you drag it to Applications, which does attach
one. Both behaviours are normal; neither means a damaged download.
What you supply
No package has a key. After installing, open Settings → Backend and either paste a CodeFlying key
(sign up from the link in that panel) or switch to on-device and give it an OpenAI-compatible
endpoint and model.
For voice, a key alone may not be enough. If you use Volcengine, the service must be an entitled
resource, not just an opened one — see
docs/getting-started.md. Android can also use the system voice instead, and
without any voice it is silent but still works on screen.
Would you rather not configure keys at all? Use the one-click page instead —
codeflying.app/andee, or
codeflying.net/andee in mainland China. Describe what you want,
tap the build button, and you get a package with the tokens already configured plus a QR code to
scan. Starting is free and the free token allowance is finite — once it runs out, tokens must be
bought. It is a separate service with no direct relationship to this project, with its own terms,
pricing and data handling. Note it produces an Android APK, not a Mac app — macOS packages can
only be built on macOS.
Install from source
git clone https://github.com/kuafuai/andee
cd andee
git checkout v0.14.2
# one variant, by hand:
./gradlew :app:assembleDebug -PdefaultLang=en -PcodeflyingDomain=myapp.codeflying.app -PcodeflyingSsl=true
# or all ten files, the way the release was built:
bash ~/.workbuddy/skills/andee-release-ship/scripts/build_variants.sh-PcodeflyingKey= and -PgoogleVoiceKey= are refused on purpose — a published package carries no
credential.
There is a getting-started guide: docs/getting-started.md, and
docs/getting-started.zh-CN.md in Chinese.
Upgrading: versionCode moves 28 → 29, so the Android package installs over v0.14.1 normally. The
Mac app's version string follows the same number. Existing memory, notebook, todos and chat history are
untouched. Nothing changes for you on Android — this release is a Mac fix, and the Android packages are
rebuilt only so the version number stays in step. On the Mac, whether screenshots work on an XDR display is
the thing to check.
Full Changelog: v0.14.1...v0.14.2