Data Processing Agreement
Last Updated: October 23, 2026
HIPAA Disclosure Included
1. Roles
Customers act as Data Controllers. AIEmployee.com acts as a Data Processor.
2. Processing Scope
We process data only to provide and support the Services.
Conversation recordings
Recording is off unless the Customer turns it on for an agent. When it is on, we store an audio recording of each browser conversation and phone call with that agent (caller and agent on separate channels; no camera video or screen shares), play the agent's spoken recording notice at the start of each recorded conversation, make recordings available only to the Customer's users permitted to view conversations, and delete them 90 days after they are saved or sooner when the Customer deletes them. The Customer decides whether to record and is responsible for providing any notice and obtaining any consent that applicable law requires from the people it records.
Cloned voices
Voice samples uploaded to create a cloned voice are processed to create a voice used only for the Customer's agents. Before a sample is accepted, the Customer's user must confirm that they are the speaker or hold the speaker's written permission, and we keep a record of that confirmation. The Customer is responsible for that permission being valid.
3. Security Measures
We maintain access controls, monitoring, and reasonable safeguards appropriate to the nature of the data.
4. Subprocessors
We use the subprocessors below to provide the Services. Each entry lists what the subprocessor does for us and the categories of Customer data it receives. Location is shown where our configuration sets one. Other subprocessors may process data in the United States or other countries where they operate.
Hosting and infrastructure
| Provider | What it does for us | Data it receives | Location |
|---|---|---|---|
| Convex | Database, file storage and backend for the Services. | All data stored in the Services, including account details, agent settings, transcripts and messages, conversation recordings a business chooses to keep, customer memory, knowledge-base files, uploaded images and files, voice-clone consent records, and billing and usage records. | United States |
| Vercel | Hosts our websites, web app and API, and runs the AI employee service. Its AI Gateway relays requests to AI model providers, including xAI, Anthropic, Google and ByteDance. | Data sent through our websites and web app, including account details, messages and uploaded files, IP addresses and request details, and the content we send to AI models through the AI Gateway. | Global edge network |
| LiveKit | Real-time audio and video network for browser voice sessions and phone calls. It also hosts the software that runs our voice agents. | Live audio, video, screen shares and chat messages from voice sessions and calls, and the transcripts, prompts and session details our voice agents work with. When session recording is turned on in our LiveKit account, LiveKit also stores session audio, transcripts and diagnostic traces. | Multiple regions, including the United States and the European Union |
| Cloudflare | Network and security proxy in front of our web app, custom domains for agencies, and hosting for Spaces, the apps that AI employees build. | Traffic to our web app, agencies' custom domains and Spaces, including IP addresses, request details and the content of pages and requests, and the content of published Spaces. | Not specified |
| Railway | Hosts our own document converter, our media editing service and a connector used by some older GoHighLevel connections. | Knowledge-base files while they are converted to text, images, video and audio while they are edited in Content Studio, and CRM data passing through older GoHighLevel connections. | Not specified |
| Upstash | Rate limiting. | IP addresses and account identifiers used to count requests. | Not specified |
AI models, speech and avatars
| Provider | What it does for us | Data it receives | Location |
|---|---|---|---|
| OpenAI | AI models that answer phone calls, run Lightning Voice sessions and the voice agent builder, summarize conversations, extract customer memory, index knowledge-base documents, transcribe recordings, and generate images and speech for AI employees. | Transcripts and messages, live audio in Lightning Voice sessions, knowledge-base text, customer memory, audio and video files sent for transcription such as Teach recordings, and prompts for generated images and speech. | Not specified |
| Google (Gemini API) | AI models for Premium Voice agents, conversation summaries and titles, prompt suggestions, filling CRM fields from conversations, avatar image generation, and Content Studio images and plans. When avatar image moderation is enabled, it also checks custom avatar reference images for safety. | Transcripts and messages, images from cameras, screen shares and files shared in Premium Voice sessions, website content used for prompt suggestions, and Content Studio prompts and reference images. For avatar safety checks it receives only normalized image pixels and fixed safety instructions, without filenames or tenant, user or provider IDs. | Not specified |
| xAI | Default AI model for AI employee chat and tasks, reached through Vercel's AI Gateway. | AI employee messages, task instructions, attached files, and results from the tools and apps the AI employee uses. | Not specified |
| Anthropic | AI models for AI employee specialist agents and tasks set to Max effort, and for the on-screen visuals voice agents create, reached through Vercel's AI Gateway. | Messages, task instructions and files handled by specialist agents or Max-effort tasks, and the conversation details a voice agent puts in a visual. | Not specified |
| TypeSafe AI | Quick yes-or-no and multiple-choice decisions that help route AI employee work and label conversations: which teammate suits a task, whether a group-chat message needs a reply, whether a check-in has anything useful to say, how a question to a voice agent's knowledge base should be searched, and a finished call's outcome, lead strength, whether a person should follow up and whether the caller asked not to be contacted. Reached through Vercel's AI Gateway.Only when we have turned this service on, for all workspaces or for selected ones. | Task titles and instructions, check-in signals, recent group-chat messages, the names and descriptions of the workspace's AI teammates and skills, questions voice agents look up in a knowledge base, transcripts of finished web and phone conversations, and text messages people send to a business's phone numbers. | Not specified |
| AssemblyAI | Speech-to-text for Premium Voice sessions, phone calls and the voice agent builder. | Live audio of what people say to an agent. | Not specified |
| Inworld | Text-to-speech voices and voice cloning. | The text of agent replies to be spoken, and voice recordings uploaded to create a cloned voice. | Not specified |
| LiveAvatar (HeyGen) | Premium video avatars for voice agents. | The agent's spoken replies as audio, and the details needed to connect the avatar to the session. | Not specified |
| bitHuman | Basic video avatars. | For avatars drawn in bitHuman's cloud, the avatar image and the agent's spoken replies as audio. For avatars drawn on our servers, license checks and usage reports. | Not specified |
| Moondream | Describes camera and screen-share images during Lightning Voice sessions so the agent can respond to what it sees.Only when someone shares video in a Lightning Voice session. | Still images from the user's camera or shared screen. | Not specified |
| fal | Image and video generation in Content Studio. | Prompts and any reference images, video or audio the customer provides, which may show real people. | Not specified |
| Replicate | Image and video generation in Content Studio. | Prompts and any reference images, video or audio the customer provides, which may show real people. | Not specified |
Knowledge, memory and web content
| Provider | What it does for us | Data it receives | Location |
|---|---|---|---|
| Weaviate | Search index for knowledge bases. | Text passages from knowledge-base documents and websites, their search vectors, and search queries. | Not specified |
| Voyage AI | Turns knowledge-base text into search vectors and ranks search results. | Text passages from knowledge-base documents and websites, and search queries. | Not specified |
| Supermemory | Managed customer memory and knowledge-base search, including connections to Google Drive, Notion and OneDrive.Only when we have turned this service on, for all workspaces or for selected ones. | Transcripts, facts remembered about customers, knowledge-base documents and website pages, and files from connected document sources. | Not specified |
| Firecrawl | Reads websites that customers point us to and runs web searches for AI employees. | Website addresses, the public content of those pages, and search queries. | Not specified |
| ThumbmarkJS | Open-source code in the chat widget that recognizes a returning browser for customer memory.Only when customer memory is on for the agent and, in ask-first mode, after the visitor opts in. | By default, a random sample of about 1 in 10,000 device fingerprints, the browser details used to calculate them, and the page path, sent to its developer. | Not specified |
Communications, integrations and publishing
| Provider | What it does for us | Data it receives | Location |
|---|---|---|---|
| Twilio | Phone numbers, phone calls and text messages. | Phone numbers, call audio and call details, text messages sent to a business's number, and business identity and address details or documents that some countries require before issuing a number. | Not specified |
| Resend | Sends our emails, such as invitations, notifications and alerts. | Recipient names and email addresses, email content, which can include lead details in new-lead notifications, and delivery status. | Not specified |
| Pica (One) | Connects apps a customer chooses to link, such as Google Drive, calendars, CRMs and email, and relays requests to them. | Authorization for connected apps and the data exchanged with them, such as contacts, calendar events, messages and files. | Not specified |
| Zernio | Publishes social media posts for customers who connect social accounts. | Connected social account details and the text, images and video in posts. | Not specified |
Accounts, payments and monitoring
| Provider | What it does for us | Data it receives | Location |
|---|---|---|---|
| Clerk | Sign-in and account management. | Names, email addresses, sign-in details, and IP address and device details used to secure sessions. | Not specified |
| Stripe | Payments, subscriptions, credit purchases and agency payouts. | Billing names and email addresses, payment card and bank details entered directly with Stripe, and transaction history. Agencies that receive payouts give Stripe their business and identity details. | Not specified |
| Sentry | Error monitoring. | Error reports, which can include the signed-in user's account ID and email address, browser and device details, and the address of the page or request that failed. | United States |
| Langfuse | Monitoring of voice agent quality and performance. | Traces of voice agent sessions, which can include transcripts, prompts, agent replies and tool results. | Not specified |
Customers can also connect services they use with their own accounts, such as GoHighLevel, calendars, Slack, Discord, Telegram, WhatsApp, Linear, their own tool servers and webhooks, or their own Twilio account. We send data to those services on the Customer's instructions, under the Customer's own terms with them.
Avatar image processing
When avatar image moderation is enabled, Google checks custom avatar reference images. Data category: normalized image pixels and fixed policy text, without filename, tenant, user, or avatar-provider identifiers. Purpose: image safety moderation before avatar processing. Images that are not approved are not kept.
Create-from-image avatars were retired in September 2026. Until then, that feature sent the normalized source image and a non-PII label to Protoface for avatar creation, operation, security, maintenance, and permitted service improvement. Images in the manual reference flow are moderated but are not automatically submitted for avatar creation.
We will update this list when we add or replace a subprocessor, and change the date at the top of this page when we do. Questions about subprocessors can be sent to [email protected].
5. Data Breach Notification
We will notify customers without undue delay upon discovery of a qualifying data breach.
6. Data Deletion
Upon termination, data may be deleted or anonymized unless retention is legally required.
Conversation recordings are deleted automatically 90 days after they are saved. A cloned voice deleted in the dashboard can no longer be used by the Customer's agents; removal of the voice from our text-to-speech provider is available on request. Voice consent records are kept after the voice is deleted.
Raw uploads are deleted after finalization succeeds or fails. Normalized sources follow authenticated owned deletion under the lifecycle below. Manual-reference normalized images for unpaid drafts are retained for seven days after binding; replacement, abandonment, cancellation, and completed fulfillment queue authenticated owned deletion. Checkout adds seven-day protection against payment and deletion races. Paid or in-progress requests are not deleted by stale-unpaid cleanup. Failed deletion will retry with exponential backoff starting at five minutes and capped at 24 hours. An assigned avatar that adopts the same image receives ownership by transfer instead of deletion. Limited consent, audit, and deletion-tombstone records may be retained where needed for compliance, security, or replay prevention.
7. HIPAA Disclaimer
AIEmployee.com is not a covered entity under HIPAA. The Services are not intended to process Protected Health Information unless a separate Business Associate Agreement is executed in advance.
Customers are responsible for ensuring PHI is not submitted without authorization.