Trust center
Security that’s built in, and honest about what’s next.
Live agents talk to your customers and learners, so we take protecting those conversations seriously. Here’s how Somebody works today.
How data flows
Three parts, clearly separated.
1 · The visitor’s browser
The agent’s face is rendered on the visitor’s device. The microphone is used only after they start a conversation.
2 · The realtime AI
Speech is processed by the agent’s AI provider: our included AI, or your own OpenAI, Gemini or Azure account.
3 · Your Somebody organization
Agent settings, knowledge, version history and transcripts, isolated to your organization.
Where AI is used
Every AI feature, and whose AI runs it.
Conversations and grading can run on your own provider account. Dashboard tools run on Somebody’s AI, through the providers on our subprocessor list, none of which may train on your data.
| Feature | What the AI does | Whose AI runs it | Your own key |
|---|---|---|---|
| Live conversations | Understands what the visitor says, writes the reply, speaks it and transcribes the conversation. | Included AI (OpenAI or Google Gemini), or your own keys for OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, ElevenLabs or Cartesia. | Yes |
| Roleplay grading | Scores a finished roleplay against your rubric and writes the feedback. | The agent’s own key when it has one. Agents on included AI are graded by Somebody’s AI. | Yes |
| Prompt assistant | Helps you draft and improve agent instructions, and simulates test conversations. | Somebody’s AI. | No |
| Resource summaries | Summarizes documents you upload so you can see what the agent will know. | Somebody’s AI. | No |
| Avatar portraits | Creates and edits AI portraits in the avatar studio. | Somebody’s AI. | No |
| 3D face builds | Turns a portrait into the 3D face the agent uses. | Somebody’s own build servers. No third-party AI provider. | No |
On your own key, Somebody’s servers do not intercept or retain conversation content. Session transcripts are stored on our servers and are optional: turn them off per agent and the conversation text isn’t kept.
Practices
What’s in place today.
Organization isolation
Every agent, resource, transcript and key belongs to one organization. Requests for another organization’s data are refused.
Write-only provider keys
Your OpenAI, Gemini or Azure keys can be added and deleted, but never read back through the dashboard.
Origin-locked embeds
Publishable keys only work from the domains you allow. Secret keys can expire and be revoked at any time.
Hashed credentials
API keys are stored hashed, and can be rotated or revoked from the app.
Transcripts you control
Turn transcript storage on or off per agent, and delete conversations whenever you need to.
Consent-aware recording
Optional recordings are made in the visitor’s browser, and your site can require consent first.
Change history
Every save is recorded as a snapshot you can compare and restore, and changes are logged.
Roles
Admins manage billing, keys and members. Members build and run agents.
Clean exports
SCORM, cmi5 and xAPI packages and transcript exports never contain your API keys.
Enterprise
Going through procurement? We’ll help.
Enterprise customers get a security review with our team, including questionnaires, an architecture walkthrough and contract terms that fit your policies.
Your AI terms
Bring your own provider agreements and data terms.
Data deletion
Delete transcripts and agents whenever you need to.
Payments by Stripe
Card data never touches our servers.
Open roadmap
We’ll share where our security programme is heading.
Where is conversation data processed?
Audio is processed by the realtime AI provider for that agent: our included AI, or your own OpenAI, Google Gemini or Azure OpenAI account. If you bring your own keys, the conversation runs under your agreement with that provider. Agent settings, knowledge and transcripts are stored by Somebody.
Can we stop transcripts being stored?
Yes. Transcript storage is a per-agent setting, and it’s enforced on our servers, not just in the interface. Admins can also delete transcripts at any time.
Do you train AI models on our data?
No. We don’t use your agents, knowledge or conversations to train AI models.
Who processes payments?
Payments are handled by Stripe. Card details go directly to Stripe and never touch our servers.
Do you hold security certifications?
We’re an early-stage company and don’t claim certifications we haven’t earned. Enterprise customers get a full security review with our team, and we’ll share our current controls and roadmap openly.
How do I report a vulnerability?
Email security@somebodyai.co with the details. We’ll acknowledge your report and keep you updated while we investigate.
Questions about security?
Ask our agent, or book time with the team for a detailed walkthrough.