Governable agents, on the record
The hard part isn’t building agents — it’s keeping them accountable. Avantis AI logs what agents do, keeps a person in the loop, and enforces access and isolation at the engine, not the prompt.
Governance you can show an auditor
Shipped and in use today — the substrate behind a compliance story.
Complete activity log
Every instruction, tool call, decision and approval is recorded with a request/run correlation id, so any action can be traced end to end.
Human approvals (HITL)
Cross a threshold and the action pauses for a person to approve — nothing consequential runs without sign-off.
Role-based access
People and agents occupy positions with roles and visibility scopes (RBAC + attribute rules); each reaches only the tools and data its role allows.
Tenant isolation
Every workspace is walled off under a default-deny policy — your data never crosses into another tenant.
Egress & injection defense
Outbound calls are filtered against an allow-list; after an agent reads untrusted content, what it tries next is held for review — aligned with the NSA/partners MCP security guidance.
Your keys, your data
Bring your own model key and credentials; secrets are encrypted at rest and never shown to the model. Spend budgets cap what each agent can run.
Standards & certifications
We’re in free beta. The controls above are real and in use; formal certifications are not yet in place. Here’s exactly where each stands — we’d rather say it plainly than imply otherwise.
- PlannedFull audit-trail exportActivity is logged today; self-serve export for auditors is on the roadmap.
- PlannedSOC 2Not yet in place — we’re in free beta and won’t imply otherwise.
- ExploringISO 42001 (AI management)Our controls map toward it; formal alignment is a roadmap goal.
- ExploringEU AI Act readinessHuman oversight, logging and risk controls are built in; a formal readiness statement is planned.
Data handling is covered in our privacy policy. Have a due-diligence or compliance question? Email [email protected].
Put agents to work — accountably
Start with one agent, with logging, approvals and isolation on from day one.