Enterprise · Security
Security Compliance and Data Control
How Hamsa keeps projects apart, controls who sees what, protects credentials and recordings, and where conversation data goes. For certifications and security questionnaires, talk to our team.
- SOC 2 Type II attested
- Isolated projects
- Owner, Admin and User roles
- Public and secret keys
- Encrypted secrets
- HTTPS-only webhooks

SOC 2 Type II
Attested
Access control
Identity and access control for conversational AI
Projects keep work apart
Each project has its own agents, knowledge base, tools, phone numbers, call history, campaigns, secrets, voices and members. Use separate projects for separate teams or environments.
Roles and permissions
Members are an Owner, an Admin or a User, and each member can be given custom permissions. Permission changes apply within a minute.
Public and secret keys
A public key sits in your page or app: it can only start calls, it works in one environment, and it can be limited to your domains. A secret key stays on your server, is shown once, and carries only the scopes you give it.
Secrets referenced by name
Credentials for your systems are stored as encrypted project secrets. Tools refer to them by name, so they never sit in a prompt.
Account sign-in
Passwords need at least 8 characters with upper and lower case, a number and a symbol. Members can also sign in with Google or Microsoft.
Webhooks over HTTPS
Call events go only to HTTPS endpoints, with bearer token authentication. Self-signed certificates are not accepted.
Data handling
Enterprise voice AI privacy: what happens to call data
Encryption
Hamsa’s privacy policy commits to encryption in transit and at rest for sensitive data, with access limited to authorised staff on a least-privilege basis.
Recordings and transcripts
Calls are recorded and transcribed for review. Access to call history is set by role, and recordings are reached through signed URLs.
The language model
Hamsa’s own models handle speech. Agent replies come from a language model you set per agent: OpenAI, Gemini, Groq, DeepMyst or any OpenAI-compatible endpoint, with Gemini by default. Conversation text goes to the provider you choose.
Your content
Under Hamsa’s terms you keep all rights in your content. Hamsa processes it only to provide and secure the service, and deletes or returns it when the agreement ends.
Secure setup
Setting up voice AI security for your organisation
The controls above, in the order a security team usually applies them.
- 01
Choose where it runs
Use Hamsa Cloud, or deploy in your own private cloud or on-premise on the Enterprise plan, depending on where your policy says call data may live.
- 02
Choose the language model
Pick, for each agent, a provider your data policy allows, or point the agent at an OpenAI-compatible endpoint of your choice.
- 03
Separate projects and assign roles
Create a project per team or environment, then give each member the role and permissions their work needs.
- 04
Issue keys and store secrets
Give each website or app a public key, and each server a secret key with only the scopes it needs. Store system credentials as secrets, not in prompts.
- 05
Decide who monitors calls
Live listen-in is silent and not available to the User role. Make sure monitoring follows the laws that apply to you, for example by telling callers that calls may be monitored.
Controls
Security controls at a glance
Hamsa is SOC 2 Type II attested. Ask our team about the other standards your review requires.
| Control | Detail |
|---|---|
| Certification | SOC 2 Type II (AICPA SOC for Service Organizations) |
| Project isolation | Agents, knowledge base, tools, phone numbers, call history, campaigns, secrets, voices, members |
| Roles | Owner, Admin, User, with custom permissions per member |
| Live monitoring | Silent listen-in; not available to the User role |
| API keys | Public keys for pages and apps (one environment, allowed domains, start calls only); secret keys for servers (shown once, scopes) |
| Secrets | Encrypted; referenced by name in tools |
| Webhooks | HTTPS only, bearer token, no self-signed certificates |
| Recordings | Reached through signed URLs |
| Sign-in | Password policy (8+ characters, mixed case, number, symbol); Google or Microsoft |
| Language model | Set per agent; Gemini by default; any OpenAI-compatible endpoint |
| Deployment | Hamsa Cloud; private cloud or on-premise on the Enterprise plan |
FAQ
Questions about voice AI security
Which certifications does Hamsa hold?
+
Hamsa is SOC 2 Type II attested. For the report, or for other standards, ask our team: procurement reviews and security questionnaires are handled by people in the region, during your working hours, so you get answers for the standards your review actually requires.
Is our data encrypted?
+
Hamsa’s privacy policy commits to encryption in transit and at rest for sensitive data. Project secrets are encrypted, and webhooks are delivered only to HTTPS endpoints.
Does a third-party AI provider see our conversations?
+
Speech recognition and voices are Hamsa’s own models. The language model that writes an agent’s replies is configurable per agent: OpenAI, Gemini, Groq, DeepMyst or any OpenAI-compatible endpoint, with Gemini as the default. Conversation text is sent to the provider you configure, so choose one your data policy allows.
Who can see call recordings and transcripts?
+
Access to call history is controlled by role, and recordings are reached through signed URLs. Live monitoring is not available to the User role. Monitoring is silent, so make sure your use of it follows the laws that apply to you.
How are API keys and credentials protected?
+
Websites and apps use a public key, which can only start calls in one environment and can be limited to your domains, so finding it in a page reaches nothing else. Servers use a secret key: shown once, refused from a browser, and limited to the scopes you give it. Credentials for your own systems are stored as encrypted project secrets and referenced by name, so they never appear in a prompt.
Who owns the content we put into Hamsa?
+
You do. Under Hamsa’s terms you keep all rights in your content, and Hamsa processes it only to provide and secure the service. When the agreement ends, Hamsa deletes or returns it.
Can we run Hamsa in our own environment?
+
Yes, on the Enterprise plan. Private cloud deploys the full stack inside your own VPC or dedicated tenancy, under your encryption keys and network rules. On-premise runs models, storage and logs on your own infrastructure.
Keep reading
Enterprise · Deployment
Cloud Private Cloud and On Premise Deployment
Hamsa Cloud, private cloud in your own VPC, or on-premise on your infrastructure, with the language model set per agent.
Read →Enterprise · Services
Enterprise AI Implementation Services
Solution engineers take Arabic voice agents through discovery, flow design, integration and go-live, with support in your region and time zone.
Read →Audio Analysis · Call analytics
Call Analytics and Quality Assurance
Call history, transcripts, outcomes, dashboards and live listen-in for every call your Hamsa agents handle.
Read →


