Privacy
Privacy Policy.
In short, we collect your account details, information about how you use Farero, and anything you choose to send us as feedback. Usage analytics help us decide what to improve, and you can turn them off. They never include your prompts, code or file contents.
Your workspaces and chat history live on your computer. Your agents send requests to the model providers you use. If you connect ChatGPT, selected chats and approved prompts pass through farero.ai to ChatGPT. Optional AI and voice features send the information they need to their providers. Here is what each part involves and how you can control it.
Last updated: 29 October 2026. This policy covers the Farero application, farero.ai and the Farero plugin for ChatGPT.
Who runs Farero
Adluna AS is responsible for the personal data described in this policy. Our organization number is 927 043 297 and our address is Waldemar Thranes gate 32C, 0171 Oslo, Norway. To ask about your data or request access, correction or deletion, use the Send feedback form in the website footer and include an address where we can reply.
Your account and connected computers
When you sign in with Google or GitHub, we receive your name, email address, avatar link and that provider's account identifier. We use them to identify your Farero account and connect your computers. Google or GitHub processes the sign-in under its own privacy policy. Signing in to Farero uses separate authorization from access to your repositories and coding-provider accounts.
farero.ai stores your account profile and an installation record containing an installation identifier, platform, Farero version, creation and last-seen times, and a hashed installation credential. Your computer stores the linked profile and installation credential locally. Devices opening the same Farero installation, including a phone, share that installation's account and access.
When you connect ChatGPT, we also process the chosen computer, permission scopes, client and authorization identifiers, redirect information, credential hashes and expiry times. We use these records to verify the connection, route requests, renew authorized access and revoke it when disconnected. We use Cloudflare to host the website, account database and ChatGPT relay.
Workspaces, coding agents and local storage
Your computer stores working files, repository and branch information, chat messages, saved agent reasoning and tool activity, attachments, prompt queues, delivery receipts, settings and local diagnostic records. Imported provider conversations are also stored locally when you choose to import them. Browsers and the desktop app keep device preferences and cached application data.
Your coding provider receives the prompts, files, tool results and other context its agent uses for your task. The agent can also send information to Git hosts, MCP servers and other tools you authorize it to use. These recipients depend on your selected provider, account and task. Their own policies and account settings govern their processing and retention.
Opening Farero remotely sends application content to your connected device over the connection you use, such as your Tailscale network. Updates and artwork downloads also contact the services that deliver those resources.
Connecting Farero to ChatGPT
The ChatGPT connection is optional. You enable ChatGPT connection in Farero Settings, sign in through farero.ai and choose a computer during authorization. The read permission allows the plugin to query native workspaces and chats on that computer, including archived native chats. A filter on an individual query narrows its results; authorization is not limited to one named chat or workspace. Imported provider history is outside this plugin's tools.
ChatGPT sends the requested tool name, arguments and authorization credential to farero.ai. The relay forwards the request to your connected computer and returns its result to ChatGPT. Cloudflare operates the relay infrastructure and OpenAI receives the returned content. Selected content is processed by the relay while it is forwarded, even though the original workspace and history remain on your computer.
The current tools process these inputs and return these categories of information:
| Tool | Information received | Information returned to ChatGPT |
|---|---|---|
workspace_overview | Optional repository, status, date and visibility filters; page cursors; a selected brief identifier. | Workspace and chat names, identifiers, activity and status; waiting questions; saved attention briefs, quotations and suggested replies; supporting passages when requested. |
list_workspaces | An optional agent-status filter. | Workspace names and identifiers, repository and branch names, models, roles, pull-request information, task status, queued or parked prompt previews and operational warnings. |
search_chats | Search words or a phrase, optional repository or workspace and date filters, ordering and result limit. | Matching workspace or chat names, identifiers, tags, activity information and short matching excerpts. Excerpts can come from user messages, assistant replies or saved agent reasoning. |
list_chats | A workspace identifier and optional choices to include helper chats. | Chat names and identifiers, models and status, context usage, recent-message previews, waiting-task descriptions and questions or approvals awaiting a response. |
read_chat | A chat identifier, optional message cursor and limits controlling the passage. | A bounded passage of saved conversation messages. This tool excludes agent reasoning and tool activity; search results and overview evidence have different content. |
send_prompt | The chosen workspace and chat identifiers, exact prompt text, a request identifier and optional queue mode. | An acceptance or delivery receipt, status, warnings or errors. The prompt is saved locally and delivered to the chosen coding agent, which can change files or use external services as part of the task. |
These results may contain personal information, confidential material or code already present in your messages and excerpts. Returned text is not automatically redacted. Review the computer and material you connect, and keep credentials and other information you do not want sent out of the connected conversations. The plugin instructs ChatGPT to obtain your approval for the exact destination and text before sending a prompt. A receipt records acceptance; it does not mean the agent has finished.
Farero's relay handles tool content in memory while forwarding the call and does not write a cloud copy of your prompt or transcript to its account database or relay storage. A pending relay call waits for up to 60 seconds before timing out. Your computer retains accepted prompts, conversation history and delivery records. ChatGPT and the coding provider can retain their own copies under their policies and your account settings.
To stop new plugin calls, turn off ChatGPT connection in Farero Settings. Farero stops routing locally and revokes the computer's ChatGPT grants when the account service is reachable. You can also disconnect the plugin in ChatGPT. Disconnecting does not delete existing conversations or cancel a prompt that the coding agent has already accepted.
Optional AI and voice features
Where available, Questions and suggested replies uses TypeSafe AI's Jev service to identify what needs your attention and select relevant passages and suggested replies. Enabling it permits selected recent native-chat excerpts and pull-request status to be sent to TypeSafe AI. Passage selection can also process a pending question and surrounding saved conversation, including saved reasoning in native-question context. Saving an API key and enabling the feature are separate steps. Turn the feature off in its Settings control to stop new requests; saved results remain on your computer.
TypeSafe AI's published privacy policy says it does not train or fine-tune models on submitted prompts or other input. Its policy describes retaining personal data for as long as reasonably necessary to provide its services or support its business purposes, with a deletion-request process.
When you use Auto to choose an agent, a routing model can receive the first prompt, submitted attachment text or images, routing rules and available model or role descriptions. This request uses a configured coding-provider account to choose a suitable agent. Choosing a specific agent avoids that Auto routing step. Routing decisions are saved with the local task.
Voice sends microphone audio, a transcript and the instructions and context needed for the conversation to OpenAI. A voice assistant can read selected workspace and chat information through tools. Delegated voice tasks can also send transcript fragments, tool results and selected chat context to an OpenAI model; previous-call history can be retrieved when requested. Farero keeps saved call transcripts and tool activity locally. These records have no automatic age-based deletion or in-app delete control in the current voice service.
If you configure telephone access, Twilio also handles the telephone call and its signaling, and OpenAI handles the voice session. Telephone service providers may receive the calling number and call metadata. Browser voice and telephone access require separate configuration; do not use them for information you do not want processed by those services. You can end a call, withdraw microphone permission and stop using or disable the configured voice connection.
Notifications and remote access
When you enable push notifications on a device, Farero stores the device's push subscription and encryption information locally. Notifications can contain a workspace name, a short assistant reply or error excerpt, a waiting-task description and a link to the relevant chat. Farero encrypts Web Push message bodies for delivery through your browser or operating-system vendor's push service. Your device can display that content on its lock screen, depending on its notification settings. You can turn notifications off for the device and manage lock-screen previews in the operating system.
Remote access can use private Tailscale access or a public Tailscale Funnel address, depending on the exposure mode you configure. The app access credential is still required. The connection sends app content to the remote browser or device you use; Tailscale processes the connection and network metadata under its own policy.
Usage analytics and feature availability
We use PostHog in the EU to understand how Farero is used and which changes help. App usage analytics are on by default. Reports include an installation identifier, an account identifier when signed in, a random browser or device identifier, timestamps, app version, platform, update channel, providers and models used, selected settings, activity and device counts, prompt-length ranges and fixed error categories. Your email is sent to PostHog when we update the account link. Reports can be linked across your installations and, where a website visit is linked during sign-in, to that browser's activity.
App usage analytics exclude prompt text, code and file contents, repository, branch and workspace names, file paths and network names. Error reports contain fixed categories and standard error types rather than the original message or stack trace. These limits describe usage analytics; the ChatGPT connection and optional AI features process content as explained in their own sections.
Turn app analytics off during setup or in Settings → General → Usage analytics. New reports stop and pending reports and retries are discarded. Re-enabling reporting does not recreate the missed activity. This switch does not delete reports already received or control analytics on the public website.
farero.ai uses PostHog for page views, downloads and feedback-submission events. Its analytics SDK can process the page URL and referrer, browser and device information, IP address and event times. It keeps a browser identifier in local storage and cookies, respects Do Not Track, and has session recording and automatic click capture disabled. Clearing browser storage removes that device's saved identifier; it does not erase earlier reports.
The app also sends an account or installation identifier to PostHog to check which features are available. These availability checks are separate from optional usage reporting and may continue when usage analytics are off.
Feedback and support
When you send feedback through farero.ai, we receive your message, its category, any reply email you provide and limited page or application context. The service files this as a private support ticket in PostHog, hosted in the EU and available to the Farero team. Earlier reports remain in the team’s private GitHub repository until deleted. Feedback works even when usage analytics are turned off. We use the message to investigate your report and the email to reply from a personal email address.
You can also send Feedback & support from the app; app reports go to the same private support tickets. App feedback may be sent without an email where the form permits it; a question requires a reply address. Technical logs are off by default. Enabling them adds a previewed summary of app version, platform, client and view category, together with up to 12 recent request categories, outcomes, status codes and timings. That summary excludes chat content, code, file paths, URLs, credentials and provider output. The text you write in the feedback message is included as written, so review it before sending. Opening the separate relay-log viewer does not attach those logs to feedback.
Hosting, security and other recipients
Cloudflare processes requests to farero.ai and hosts the account database and ChatGPT relay. This includes the content needed for a requested operation and network information such as an IP address, request time and HTTP headers. Farero's sign-in and feedback rate limits use temporary IP-derived fingerprints and request times in service memory to limit abuse.
PostHog processes the analytics, feature-availability and feedback-ticket information described above. GitHub processes GitHub sign-in and retains earlier feedback reports; Google processes Google sign-in. OpenAI processes requests and results when you use ChatGPT or an OpenAI-powered feature. TypeSafe AI processes the optional attention requests. Twilio, Tailscale and your device's push provider process the connections described above. Your chosen coding providers and connected tools process the context needed for their tasks.
We use this information to deliver the requested features, maintain account and connection security, answer support requests and measure product use. We do not use ChatGPT plugin prompts or returned chat content to train an Adluna model. Other services apply their own policies and account settings. In particular, PostHog's published policy describes product and model development using aggregated or de-identified customer content, subject to applicable contract terms and opt-out settings; that content does not include the prompt or code text excluded from Farero usage analytics.
Providers may process information outside your country. TypeSafe AI describes its services as hosted in the United States. Review the relevant policies and the settings for your account: Cloudflare, PostHog, OpenAI, TypeSafe AI, GitHub, Google, Twilio and Tailscale.
Retention
| Information | How long it is kept |
|---|---|
Local workspaces, chat history, accepted prompts, attachments, delivery records, saved attention results and voice transcripts | Kept on your computer until removed, without a general age-based expiry for saved history. Archiving a workspace and uninstalling or stopping the service preserve history and working files; they are not erasure requests. Copies in backups you control remain until you remove those backups. |
Account, identity and installation records at farero.ai | These records have no automatic age-based deletion in the current service. They remain until erased in response to a request or an administrative deletion. Signing out or revoking a credential does not erase the profile or installation record. |
ChatGPT tool content at the relay | Handled in memory during the call; the relay waits up to 60 seconds for a result. The relay does not durably store a chat-content copy. Local and recipient copies follow their own retention. |
ChatGPT connection attempts and one-time authorization codes | Connection attempts expire after 10 minutes and one-time codes after 2 minutes. Consumed records are removed during use; expired ChatGPT authorization records are swept daily. Expiry stops their use before that sweep. |
ChatGPT authorization credentials | Access credentials expire after 1 hour. Refresh credentials expire after 30 days and are replaced when used to renew access, which starts a new validity period. Disconnecting the computer deletes its stored ChatGPT codes and credential hashes; expired records are swept daily. |
Other account sign-in attempts and one-time codes | Valid for 10 minutes and 2 minutes respectively. Used records are consumed once; expired records are cleared when a new account sign-in starts. |
Usage analytics and associated profiles | Farero does not apply an automatic age-based deletion schedule to reports already sent to PostHog. They remain in our analytics project until deleted by us or expired by PostHog's retention settings. You can ask us to delete personal analytics and associated profiles. Turning reporting off stops new reports; it does not erase those already received. |
Feedback and support issues | Farero applies no automatic age-based deletion to feedback reports. New reports are kept as private PostHog Support tickets, subject to PostHog’s retention settings; earlier reports remain in the team’s private GitHub repository until deleted. You can ask the Farero team to delete a report from either store. |
Content held by your coding provider, ChatGPT, OpenAI voice, TypeSafe AI, Twilio or another service you use | Retention is governed by that service and your account settings. Farero's local storage and disconnection controls do not delete those copies. TypeSafe AI's published retention criterion is described in the optional AI section above. |
Temporary abuse-prevention fingerprints and times | Kept in service memory, with requests counted over a one-hour window. Memory is discarded when the service instance is replaced or an entry is evicted; these rate-limit maps are not a durable database. |
Cloudflare's built-in database recovery history can preserve earlier account and authorization records for up to 30 days. Deleting a live record does not immediately erase these earlier recovery states. Cloudflare also processes network and security information under its own privacy policy. Our website Worker has no configured Logpush export or tail consumer to collect a separate copy of its request logs.
Your choices and data requests
You can disconnect ChatGPT, turn off app usage analytics, stop optional AI processing, sign out of the installation and manage browser storage. Signing out removes the local account link and attempts to revoke the installation's server credential. For data already held by ChatGPT, your coding provider or another connected service, use that service's account controls and request process.
For personal data held by Adluna AS, use Send feedback in the farero.ai footer to request access, correction or deletion, or ask a question about a particular data flow. Include the email associated with your Farero account and a reply address. We may need to verify that the request is yours. Avoid including passwords, authentication codes, API keys or unrelated private chat content in a data request.
Disabling a feature changes future processing. Erasing existing local records, hosted account data, feedback or analytics is a separate action, so name the information you want removed in your request.