Governance Flow
Whatever an agent does — a lookup, an edit, an email, a payment — it moves through the same five gates.
01
Request
An agent wants to read, write, or act.
02
Permission check
Scoped to role and workspace.
03
Approval gate
Sensitive actions wait for a human.
04
Scoped action
Runs only with granted access.
05
Audit log
Who, what, when, and why — kept.
Three Pillars
Three controls decide what an agent can touch, when it must wait for you, and what gets remembered.
Permissions
Approvals
Audit
Architecture
Separation between customers is architecture, not a setting.
A dedicated database per workspace
Your records are never rows in a shared database. Export them anytime.
A sandboxed runtime per agent
Every user's agent runs in its own isolated runtime, spawned on demand.
Pinned to your region
US West or EU West, chosen at signup. That choice can't silently change.
Tool Control
Connected tools are scoped up front, gated on the way out, and logged.
Scoped connections
Every integration has explicit scopes. An agent can only call what its role allows.
External writes wait
An email, a payment, a posted invoice: anything that leaves the workspace waits for sign-off.
One audit trail
Every tool call lands in the same log: who called what, when, and what came back.
Ownership
Who owns your data, what it's used for, and whether you can leave: your call.
Own your data
Your own database, not rows in someone else's. Export it anytime.
No training on your data
Your business data is never used to train models.
Backed up automatically
Backed up automatically, so a mistake never becomes a loss.
Open customization layer
The apps Opser generates are open source. Read every line, modify it, or fork it.
Security FAQ
Yes. Agents run under the same roles and permissions model your team uses, scoped to role and workspace. They see only what they're cleared for; sensitive records and tools stay out of reach by default.
Risky or irreversible actions — sending, paying, deleting, and writes to external systems — pause for human sign-off before they run. Approvals happen in-flow, where the work is, not in a separate console.
In the region you choose at signup: US West or EU West. Your workspace's data lives there, in your own database, and that choice can't silently change.
No. Your business data is never used to train models.
Yes. Your data lives in your own database and can be exported anytime, and the audit history is exportable too. The apps and business logic Opser generates are open source, so you can read, modify, or fork them.
Access follows the same roles and permissions model as everything else in your workspace — the people you clear for it. Every entry traces an action to the user, the agent, the input, and the outcome.
Yes, architecturally. Each workspace runs on its own dedicated database, every user's agent runs in its own sandboxed runtime spawned on demand, and both are pinned to the region you chose at signup. There is no shared database and no shared execution environment.
Only the context the acting user's role is cleared for, for the task at hand. Permission checks happen before the agent reads or writes anything, and none of it is used to train models.
Yes. Every workspace's database is backed up automatically, and because the data is yours you can also export it anytime for your own copies.