harperOS holds none of your data.
harperOS builds the system and hands you the keys. The repository, the data, and every credential stay yours.
Where it runs
Ownership
Access
Guardrails
Transparency
Data protection
The system runs on your infrastructure, not ours.
Your GitHub
Your Workspace
Your machines
Nothing with us
Every rule is plain text in your own repository.
1# communication-safety2 3Customer communications require coordinator review. Draft first, never auto-send.4 5The pipeline for any outbound communication (email, Slack, Teams): produce a draft → present it in chat for review → wait for explicit approval → send only after approval.6 7"APPROVED" is the only valid approval signal. Silence, nodding, or implicit consent are not approval.8 9Never send a communication that has not been explicitly approved by the user. Never send a communication from a tool call in the same step as drafting it.You own the system, the data, and the agents.
Everything we build is licensed to you, in your own repository, for good. You can move the whole system to another provider whenever you want. Your workspace runs on its own login and its own roster — you decide who gets in.
One agent, one person.
Every agent signs in as its owner and reaches only what that person can. You grant access by name and revoke it in one step. Whatever security you already run keeps applying: your single company sign-in (SSO), two-factor login (MFA), and rules that stop sensitive data leaving (DLP). We add nothing that goes around it.
The agent proposes. A person decides.
The agent can prepare anything and finish nothing on its own. It drafts the email, your team presses send. It reads from your live systems but never changes them. Every change is staged for a person to approve, and made under that person’s own login. There is no all-powerful admin account for the agent to use, and none of your data trains any AI model.
You can read every line.
Every rule and every integration is plain text in your own repository. There is no hidden backend — your team, and your security reviewers, can read exactly what each agent does. Every action is logged.
Encrypted by your own cloud.
Your own cloud encrypts your data, in transit and at rest. It is scrambled while it travels over the internet (TLS 1.2+), and again while it’s stored (AES-256).
What your security team will ask.
SOC 2
We are working toward a SOC 2 Type 2 report. Our controls are monitored continuously in Vanta, and you can see where they stand in our Trust Center.
Your data and the models
Your data is never used to train AI models. Claude runs under your own Anthropic account, which you license and pay for directly.
Who touches your data
The vendors that touch your data are the ones you already use: your GitHub, your Google Workspace, your own systems, and Anthropic’s Claude, each under your own accounts. Our website’s vendors are listed in our privacy policy.
Documents for your review
Request our security overview and policies at security@harperos.ai, and we send what your review needs.
SOC 2.
We are actively engaged in obtaining our SOC 2. Our security controls are monitored continuously. See where they stand in our Trust Center.
Security questions go to Pierre Merzeau, our CTO and Security Officer.