Codeloud privacy policy

Last updated October 20, 2026.

Codeloud is a Windows dictation app. It transcribes audio through the provider you select and can rewrite text through a separately selected model. This policy covers the current desktop app and codeloud.dev. Existing codeloud.xyz privacy links remain available.

Audio and model requests

When you transcribe a recording or an imported file, Codeloud sends the audio to your selected speech provider. Live transcription sends audio while you are recording. Other transcription modes send the recording after capture. Longer imported recordings are split locally into audio parts before you review and start transcription.

If you enable writing or request another rewrite, Codeloud sends the relevant transcript, editing instructions, selected mode, and any approved dictionary reference to your selected writing provider. The speech provider and writing provider can be different companies. Local text replacements and text formatting run on your computer.

Provider requests go from your computer to the configured API endpoint. The current app does not relay them through a Codeloud-operated server. Supported connections include Groq, OpenAI, Deepgram, ElevenLabs, xAI, Anthropic, Google Gemini, OpenRouter, Cloudflare Workers AI, Ollama, and custom compatible endpoints. Only the services you use receive requests. Cloudflare Workers AI requests use its direct API, without Cloudflare AI Gateway.

The receiving service can also receive your API authentication, IP address, selected model, language, request timing, and other information needed to handle the request. A service such as OpenRouter can route the request to another model provider. If you configure a custom endpoint, its operator receives the request. An Ollama or other endpoint running entirely on your computer can process writing requests locally, but that does not make a separately selected cloud speech service local.

Each provider has its own privacy, retention, security, and model-training terms. Codeloud does not control those practices and does not promise that a provider retains no data or never uses it for training. Canceling a request does not retract data already sent. Changing providers affects subsequent requests.

The microphone test measures audio levels locally without sending a transcription request. Microphone access is subject to your Windows and browser-runtime permissions.

Information kept on your computer

Codeloud stores settings needed to operate the app. These include provider and model choices, custom endpoint URLs, a Cloudflare account identifier when configured, microphone selection, shortcuts, vocabulary, text replacements, writing instructions, and saved writing profiles.

Session history is enabled by default. It stores original transcripts, final text, rewrite attempts, timestamps, duration, word counts, processing times, provider and model names, request identifiers, token counts, reported or estimated costs, errors, warnings, and relevant session settings.

Imported-file records can contain the file name, size, format, a SHA-256 fingerprint, and audio-part boundaries needed to identify a retry. They do not contain the original file's absolute path. History retains up to 5,000 session records, replacing the oldest records as new ones exceed that limit. There is no automatic time-based expiration within that limit.

Saved prompts, dictionary entries, text replacements, writing profiles, and named project file selections remain on your computer until you remove or replace them. Named file selections contain relative file paths and a project-binding fingerprint.

The optional owner prerelease editor link stores a passage explicitly selected and sent from a local VS Code editor, with its relative filename or basename, language, document version and selection range. It retains one passage and its append receipt until you dismiss it. Appended text also remains in the prompt draft until changed or discarded. Pairing credentials stay in memory; restarting requires a new connection. Sending a selection to Codeloud does not itself send it to a model provider. Later explicit model actions follow the provider-request rules above.

API keys are encrypted before Codeloud saves them, using Windows-backed encryption. The native Windows app uses CurrentUser DPAPI protection. Codeloud decrypts a key in memory when it needs to authenticate a request to the corresponding configured service. Removing a saved key from Codeloud does not revoke that key with its provider.

Ordinary settings, transcripts, saved prompts, and project file selections are stored in local files without additional content encryption by Codeloud. Windows account permissions, device encryption, backups, and other software on your computer affect their protection. API-key encryption does not encrypt the rest of your history.

Audio retention and restart recovery

Persistent text recovery independent of ordinary history applies to desktop package version 1.0.4.0 and later. Earlier packages require session history to be enabled to retain transcript text across restarts. The recovery update is prepared for a subsequent Microsoft Store release.

Ordinary recordings and selected audio are held in memory for processing and recovery during the running app session. Codeloud does not add microphone recordings to a persistent audio library. Session history stores text and metadata, rather than the recording itself.

The owner development build can keep one ordinary batch microphone recording in memory after a confirmed failure with no transcript. Retry transcription sends that recording through the current model and formatting settings without temporary speech hints, and may incur another provider charge. Retry returns text for review without automatic copy or paste. Discard recording, deletion of the associated history entry, clearing History, or closing the app releases this audio. This recording is not available after a crash or restart.

Text recovery copies are saved locally even when ordinary session history is disabled. They contain available original or final transcript text and its session and usage metadata. Live partial text is checkpointed as it arrives. Recovery copies remain until you delete the entry or clear History; they are not automatically expired or replaced by newer recordings. They use the same local-file protection as ordinary transcripts, without additional content encryption. After a restart, unfinished copies are marked interrupted and can be reviewed in History. Startup does not resend audio, retry a provider request, or paste text into another application. A copy that could not be saved, words not captured, and provider output not received cannot be recovered by restarting.

Audio-file batches have a separate option, Keep this batch after closing. When you choose it, Codeloud saves the batch and remaining prepared audio locally in an encrypted queue. The queue uses AES-256-GCM, with its key protected by Windows-backed encryption. It can also contain file names, progress, results, and retry information. A restored batch requires review and an explicit start before more provider requests.

Pending or unsuccessful audio can remain in the retained queue until you complete it, remove the affected item, or clear the batch. Completed audio, canceled audio, and audio whose transcript is already available are removed from the retained batch as it advances. Clear batch removes the saved queue. The original files you selected remain where you stored them.

Project folders and coding context

Advanced project features operate on a folder you choose. Codeloud can inspect its directory structure, file names, selected text files, and Git information to support local search, file references, source review, and prompt composition. Selecting a folder does not automatically upload its contents.

File-name matching and source search run locally. Reviewed source excerpts, file trees, and diffs can be included in a prompt that you copy. Copying a reviewed prompt places that text on the Windows clipboard. Another app or service receives it if you paste or submit it there.

If you explicitly approve project speech hints, selected file names can be sent to the selected speech model. Dictionary terms can also be included in supported speech requests. Sharing dictionary terms with a writing model requires approval for that recipient. The relevant terms and reviewed filename-hint information can be retained in session history when history is enabled.

In the owner prerelease, you can also review names extracted from selected source files and approve them for one recording. Codeloud sends the approved names in supported speech hint fields, alongside audio, to the speech recipient shown in the review. This action does not send the source file bodies or add names to your persistent dictionary or writing-model sharing preferences. The recording's local history or recovery record can retain the selected names, relative source locations, file fingerprints and reviewed request information. The one-recording approval itself does not survive restart.

Clipboard, paste, and exports

Automatic copy is enabled by default. Optional automatic paste inserts results into another application. The paste helper checks foreground-window, process, focus, and clipboard information locally to reduce wrong-destination pastes. Those checks are not uploaded to AI providers.

The Windows clipboard, clipboard history or synchronization, destination applications, and files you export can retain their own copies. History, dictionary, and transcript exports are ordinary files at destinations you select. Codeloud does not encrypt those exports or control what another application does with copied text.

Diagnostics and network activity

The current desktop app has no advertising SDK, publisher analytics upload, or built-in cloud synchronization. Usage figures are calculated or recorded locally from your sessions and provider responses. Codeloud does not upload those usage records to a Codeloud analytics service.

The native app uses authenticated loopback connections on your computer to connect its interface, recording control, and background process. These local connections are separate from requests to your configured model providers.

Windows, Microsoft Store, WebView2, your chosen AI providers, and other software on your device may handle their own diagnostic or service data under their respective policies. If you share logs, screenshots, exported history, or a support message, the material you choose to share may contain personal or confidential information.

Your controls and deletion

You can stop recording and cancel processing in the app. You can disable ordinary session-history saving for future sessions, delete individual entries including their recovery copies, or clear History to remove both collections. Disabling ordinary history does not delete existing entries. In version 1.0.4.0 and later, it also does not disable transcript recovery copies.

You can clear an audio batch and remove saved prompts, project file selections, writing profiles, dictionary entries, replacement rules, and provider credentials through their respective controls. Clearing history also resets the audio batch. History deletion does not remove separately saved prompts, dictionary entries, exports, clipboard copies, original audio files, or data already received by a provider.

For provider-side access or deletion requests, use that provider's account tools or contact the provider. Codeloud cannot delete a provider's records through local history deletion. Revoke an API key through its provider if you want to prevent further use of that credential.

Local deletion updates or removes Codeloud's active files. It is not a promise of forensic erasure from storage media, operating-system recovery, backups, or copies held elsewhere. Uninstall behavior depends on the installation method and Windows settings, so do not rely on uninstalling alone to clear every copy.

Website visits

The Codeloud website is hosted through Cloudflare. Visiting it sends your IP address and HTTP request information to Cloudflare to deliver and protect the site. Cloudflare processes this information under its privacy policy. The site has no submission forms, marketing cookies, or analytics scripts. Visiting the site does not upload desktop recordings, transcripts, project files, or API keys.

Contact and changes

For privacy questions or requests about Codeloud, email support@codeloud.xyz. Do not include API keys or sensitive recordings in your message.

Changes to data handling will be reflected in this policy and its last-updated date. The public policy is available at codeloud.dev/privacy.