Assistant terms & privacy
Updated 5 November 2026. This assistant is Gagan Gehani’s AI representative. It can make mistakes and cannot make commitments on his behalf.
Public page analytics
Public portfolio pages automatically count page views and feature events using a random session ID held only in the current page’s memory. It continues across navigation within the page and resets on a full reload, restoration from browser history, or after 30 minutes without activity. It does not identify a person or recognise a returning browser.
Analytics stores the public page path, event type, timestamp, referring site’s hostname when available, country and broad device/browser categories in Cloudflare. It does not store tracking IDs in cookies or browser storage, use a fingerprint, or collect full URLs, campaign tags, raw IP addresses, raw user-agent strings, chat messages, job descriptions or audio.
Analytics preferences in the footer let you stop collection or enable it again. A nonidentifying privacy cookie remembers only an explicit refusal for up to 180 days. Your on/off choice is also saved as a plain browser preference so pages already open keep the same setting. This preference has no identifier and remains until you change it or clear the browser’s stored data. Prior refusal, Global Privacy Control, Do Not Track, detected automation and owner/testing exclusions are respected. Analytics records have a 180-day retention policy, separate from the conversation archive below. Cookie-free collection alone does not establish a universal exemption from privacy requirements.
Your recording choice
Starting a conversation or sending a message after the recording notice agrees to saving its transcript and, when you use voice, your submitted voice clips and the AI’s spoken replies. Nothing is saved in this conversation archive before that action. The microphone records only while you hold the talk control and grant browser microphone permission.
Open Details and clear Save audio and transcript before starting to use the same assistant without this website saving transcript or audio. You can also use it during an idle conversation to stop saving future turns. It does not delete earlier recordings. The choice lasts for this page session; reloading shows the recording choice again.
What is saved, and why
Gagan can privately review recorded conversations to understand visitor questions and improve the assistant. The archive contains the text, submitted voice clips, generated voice replies, timestamps, conversation and turn IDs, prompt version, selected project IDs and provider/model details. It does not record ambient microphone audio.
Archive access is restricted to gagangehani@gmail.com through Cloudflare Access. Audio is served through authenticated private endpoints; it has no public bucket URL. Transcript contents are not included in public analytics.
Retention, backups and deletion
Saved sessions have no automatic expiry and are not removed to enforce a session-count limit. They stay until Gagan deletes them. New saves pause when the archive reaches its 4 GB storage budget. Private backups may be kept on Gagan’s Mac or in his private Google Drive backup folder; backing up does not automatically delete the server copy.
To request deletion of a conversation and its backup copies, email gagangehani@gmail.com with its approximate date and time or session ID. Earlier conversations that were never stored cannot be recovered.
Providers still process unrecorded conversations
Voice input goes to Sarvam for transcription. Questions, the current job description when supplied, conversation context, portfolio evidence and relevant published website sections go to OpenAI for reasoning. OpenAI response storage is disabled for these requests; provider retention policies still apply. Reply text goes to ElevenLabs for speech. Cloudflare hosts the website and private archive. Choosing an unrecorded conversation disables this portfolio’s transcript/audio archive; each provider’s own processing and retention policies still apply.
Sarvam data policy · OpenAI data controls · ElevenLabs privacy policy · Cloudflare privacy policy
Questions and call follow-up
You can choose a time on Gagan’s Calendly page, or leave your email, question, relevant conversation context and optional Calendly link or available slots with timezone. The form lets you review or clear the context before saving. “Save for Gagan to reply” shares those details privately with Gagan so he can answer or arrange a call. You can skip the form and keep exploring.
Follow-up details are separate from audio and transcript recording: choosing not to record does not prevent you from choosing to save a follow-up. Contact details entered in this form are stored in Gagan’s owner-only dashboard and are not sent to the AI or speech providers. They have no automatic expiry; Gagan can delete them. To request deletion, email gagangehani@gmail.com.
Reply timing is Gagan’s commitment shown for his configured timezone when you submit. The site does not send emails, issue invitations or reserve a call automatically. Complete and confirm the Calendly booking to book a time. Calendly clicks and follow-up submissions are counted separately; the site has no connected booking confirmation source.
Using the assistant
Avoid sharing passwords, financial details, confidential information or other people’s sensitive information. Responses describe the portfolio’s documented work and may be incomplete. You can stop, end the conversation or leave at any time. Voice sessions last up to five minutes and share a limited site allowance.