Privacy policy

Talqing is a platform for building and running AI voice, video and text agents. It is operated by Oggnai Technologies Private Limited (CIN U62011DL2025PTC446040), 293 S/F, Western Marg, Saidulajab, New Delhi, Delhi 110030, India — “Talqing”, “we”, “us”.

This policy covers talqing.com, our documentation at docs.talqing.com, the Talqing dashboard at app.talqing.com, and the Talqing API.

Two different groups of people appear in it, and they are not treated the same way:

• Customers — the people and organisations who hold a Talqing account and build agents. For their data we are the controller, and this policy is the full description of what we do.
• End users — the people who call, message or talk to an agent that a customer has built and published. For their data we are a processor acting on that customer’s instructions. The customer decides what is collected, what the agent says, whether the call is recorded and how long any of it is kept; their own privacy notice governs it, not this one.

Account data. Sign-in is Google OAuth only — there is no password to store, and we never see one. From your Google profile we keep your email address, your Google account identifier, your display name and your profile picture URL, refreshed each time you sign in. We also keep which workspaces you belong to and your role in each, and a record of every personal access token you create — the token text itself is not stored, only the record that authorises it.

Workspace content. The agents you build and every published version of them: prompts, greetings, model choices, tools and their code, agent tasks, knowledge bases, webhook endpoints, and your workspace settings.

Session records. For every agent run we store which agent and published version handled it, the channel (voice, video or text), start and end times, duration, how it ended, latency and usage counters, and the platform fee it was charged. For calls placed over the telephone network this includes the calling and called numbers in E.164 form, and the destination of any transfer to a human.

Conversation content. The transcript of each session as text — what the end user said, what the agent replied, every tool the agent invoked and everything those tools returned, the values the agent captured during the call, and any images sent into the conversation. We strip EXIF metadata, including GPS coordinates, from every image before storing it.

Call recordings. Voice and video agents record by default. One stereo audio file per call — the caller on one channel, the agent on the other. If the agent watched a shared screen and its author asked for that to be kept, a second file holds the screen as low frame-rate video. Both live in object storage and are reachable by the workspace through links that expire after an hour. Section 8 covers recording and consent on its own.

Post-call analysis. Unless the agent’s author turns it off, the finished transcript of a voice or video call is passed once to a language model to produce a summary, optionally a judgement of whether the call succeeded, and any fields the workspace asked to extract. That model call runs on the customer’s own provider key, not ours.

What an agent remembers between calls. Each caller identity a workspace has spoken to — a phone number, a Telegram account, or an identifier the customer supplied — has a contact record that accumulates what the agent learned, so it recognises that person on the next call. It exists until the last conversation with that person is deleted.

Recipient lists. An outbound call batch holds the phone numbers a customer uploaded and whatever per-person fields they attached. An email batch holds email addresses, the message drafted for each recipient and its delivery result.

Credentials you give us. Provider API keys you bring, OAuth tokens for apps you connect, telephony account credentials, messaging bot tokens, and any secrets you store for your tools. These are encrypted at rest and are never returned to the dashboard, the API, a model or a transcript after you save them.

Knowledge base content. When you point a knowledge base at a URL we crawl that site with a headless browser — following its own links and sitemap, up to a thousand pages — and store the text we extract. We also read PDFs linked from those pages and describe a small number of images per page. That extraction runs on our own model account; see section 4.

Technical data. Our servers log IP addresses, request paths, timestamps and error traces, and emit operational metrics. Both the logging and the metrics stack are our own; neither is sent to a third-party service.

Payment data. Card details never reach us — checkout is hosted by our payment processor. We pass them your email address and name, and we keep the record of which credit pack was bought, what it cost and whether it was paid.

• Running the service — authenticating you, executing your agents, connecting calls, and sending each turn to the providers your agent is configured to use.
• Showing you your own data — transcripts, recordings, call history, analysis, usage and cost in the dashboard.
• Billing — metering platform-fee usage against your credit balance.
• Security, debugging and abuse investigation — diagnosing errors and outages, and looking into misuse of the platform.
• Support, when you ask us for it.
• Service messages — changes that affect your account, your agents or this policy.

We do not sell your data, and we do not use your conversations, recordings or knowledge bases to train models. We have no models of our own to train, and the providers your agents run on do so under your account and your settings with them rather than ours — if you have turned training off in your provider’s console, that is the setting in force. We run no advertising, and there is no advertising or analytics tracker anywhere on this site or in the dashboard.

There are two kinds of third party here, and treating them as one list would misdescribe how the platform works.

Providers you connect yourself

Talqing is strict bring-your-own-keys: no credential of ours ever backs an agent run. Every language model, speech-to-text, text-to-speech and avatar an agent runs is called with your key, on your account, and that provider bills you directly. The same is true of the carrier that carries a phone call, the apps you connect for tools, and the messaging channel you attach.

For all of these we transmit data on your instruction, and we hold no relationship with them on your behalf. Your own agreement with each provider is what governs their handling of it, including whatever you have configured there about retention and training. We cannot read or override those settings.

Which of them sees a given conversation depends entirely on how that agent is configured. An agent with no phone number never touches a carrier; an agent with no integrations never leaves its model stack.

Some models also offer built-in tools — web search, code execution, file search. Enabling one lets that model’s provider act on the conversation on their own infrastructure, under the same agreement of yours.

Our own processors

These we chose, and they act for us under our contracts. This is the complete list.

That last row is the one exception to bring-your-own-keys, so it is worth stating plainly. The CoPilots read the workspace configuration you are editing and can call the same API you can; the knowledge-base builder reads the pages you asked it to crawl. Both run on our own OpenAI account. Nothing about a live agent run — no call, no message, no task — ever does.

We disclose data outside these two lists only where the law requires it, or to protect the rights and safety of our users — and we will tell you unless we are legally barred from doing so.

Talqing runs one global control plane and two independent regions. You choose a region by which API you call. Nothing is copied or replicated between them, and no screen in the product merges them.

Beyond our own footprint, running an agent generally means an international transfer, because the providers you connect are wherever you have chosen to run them. For transfers to our processors we rely on the standard contractual clauses in their terms; for the providers you connect yourself, it is your own agreement with them that carries the transfer.

By default we keep call content indefinitely. A workspace can set a retention policy — any number of days between 1 and 3650 — and when a call reaches that age its content is deleted while the record of the call survives, because an invoice has to keep resolving. Only an admin can set it. Each call’s deadline is fixed when the call ends, so changing the policy later does not move a deadline that already exists, in either direction.

The same erasure runs immediately when a customer deletes a call themselves. It is not an audio feature: a transcript is personal data in exactly the way the recording is, and it is the copy that travels into analysis, webhooks and the model provider.

Customers can delete a call and its content, or just its audio, from the dashboard or the API, and can delete agents, tools, knowledge bases, secrets, provider keys, integrations and webhooks at any time. A text conversation has no delete of its own today; its content goes when the sessions inside it are erased, and the contact record goes with the last conversation.

• Traffic between your browser, our API and our infrastructure is encrypted with TLS.
• Provider keys, OAuth tokens and stored secrets are encrypted at rest with a key held outside the database, and are write-only over the API — no endpoint, model or support path returns one.
• Recordings and attachments sit in private buckets. Every link to one is signed, expires within an hour, and is scoped to a single object.
• Every workspace's data carries its identifier and every query is scoped to the workspace making it. Storage operations additionally refuse to touch an object outside the workspace's own prefix.
• Databases, queues and caches listen on no public interface. The signing key that mints sessions and tokens exists only on the control plane, so no region can forge another region's credentials.
• Requests that agents make outward — HTTP tools, code tools, webhooks — are blocked from reaching private, loopback and internal network addresses.
• One honest exception: the leg of a phone call that crosses the public telephone network is carried by the carrier in the clear, as PSTN telephony everywhere is. It is not encrypted end to end and we do not claim it is.

This is the section most likely to matter to you, whichever side of an agent you are on.

• Voice and video agents record by default. One setting on the agent turns it off. Recording follows whichever agent is speaking, so an agent with it off records silence for as long as it holds the call.
• Whether callers are told is a separate setting, and it is off by default. Turned on, the agent must speak a notice in its greeting — checked when the agent is published, so it cannot be switched on and then quietly dropped.
• Recording can always be stopped. Whenever a call is being recorded, the agent carries a tool that stops it, and is instructed to reach for that tool as soon as a caller objects or asks for the recording to be deleted. Stopping discards the whole file, not merely what would have followed. That right does not depend on anyone having announced anything first.
• Screen recording is off by default and only applies where a caller has chosen to share their screen with the agent.

If you hold a Talqing account. Depending on where you live you may have the right to access your data, correct it, delete it, export it, object to processing, or withdraw consent. Most of these you can exercise yourself: agents, tools, knowledge bases, connections, calls and recordings are all deletable in the dashboard, and everything the dashboard shows you is readable over the API in a portable form. For anything else — including deleting the account itself — write to hello@talqing.com. We will respond within 30 days.

If you spoke to an agent someone built here. Your rights are against that business, not against us; they control what was collected and they can delete it. Contact them first. If you cannot reach them, write to hello@talqing.com and we will pass the request on and tell you that we have.

If you are unhappy with how a request was handled you may complain to your data protection authority. In India, that is the Data Protection Board.

We set exactly one cookie, talqing_session. It keeps you signed in after you authenticate with Google. It is HTTP-only, sent only over HTTPS, scoped to talqing.com so it reaches each region’s API, and it lasts 30 days. It is strictly necessary, it is not used for tracking, and there is no analytics, advertising or third-party cookie anywhere on this site. That is why there is no cookie banner.

The dashboard also keeps a little state in your browser’s local storage — which region you last used, which workspaces you have seen, and the width of a panel you dragged. It never leaves your browser.

Two things reach out from the page itself. Fonts are served from our own domain rather than from a font CDN. And before you sign in, the home and login pages ask our DNS provider’s edge which country the request came from, so they can quote prices in your currency and offer you the nearer region — that is one request per page, carrying your IP address, and it sets nothing.

Talqing’s use of information received from Google APIs adheres to the Google API Services User Data Policy, including the Limited Use requirements. We touch Google in two places and nowhere else.

Sign-in. We request openid, email and profile, and use them only to identify you and to show your name and picture to the other members of your workspace.

Google Calendar, and only if you connect it. We request the narrowest scopes that let an agent do scheduling work: your list of calendars and their events, read-only; free/busy information; and write access limited to events on calendars you own — deliberately not the broader scope that would reach every calendar shared with you. We request no Gmail, Drive or Contacts scope, and there is no way to add one from the dashboard.

Calendar data reached this way is used only to run the tools your agent calls. It is not transferred to anyone except as needed to provide that feature, it is never used for advertising, and no human at Talqing reads it — except with your explicit permission, where it is necessary for security, or where the law requires it. Disconnecting the integration in the dashboard revokes the token; you can also revoke it yourself at myaccount.google.com/permissions.

Talqing is a business product and is not directed at children. We do not knowingly collect data from anyone under 18. If you believe a child has given us data, write to hello@talqing.com and we will delete it.

We update this page when what we do changes, and revise the date at the top. If a change materially affects how we handle your data, we will tell account holders by email to the address on the account, or in the dashboard, before it takes effect.

For anything in this policy — a question, an access or deletion request, or a complaint — write to hello@talqing.com. Under India’s data protection rules that address also reaches our grievance officer, who will acknowledge a grievance within the period the rules require.

Oggnai Technologies Private Limited
293 S/F, Western Marg, Saidulajab, New Delhi, Delhi 110030, India
CIN U62011DL2025PTC446040
hello@talqing.com
+91 99101 80529

The engineering detail behind sections 6 and 8 — exactly what an erasure deletes, and how the recording settings behave — is documented at docs.talqing.com.