Human command
- The operator can inspect what the system did and why.
- Risky actions wait for approval instead of being hidden in automation.
- Private memory and workflow history stay under local control by default.
Security model
Supervised automation by default.
Axiom packages Hermes Agent for recurring work while keeping risky capabilities behind explicit approvals and visible logs. The product principle is humanistic AI: useful autonomy with human command, local control, and inspectable behavior.
Default boundaries
- Local-only dashboard by default.
- No anonymous messaging access.
- Read-only workflows first.
- Approval required for write, send, shell, delete, external contact, and purchase actions.
- Config snapshots before updates.
Audit trail
- Tool calls, scheduled runs, approval requests, errors, retries, and model routes are logged.
- Logs are grouped by workflow and connector.
- Exports include config, memory, templates, and log metadata.
Beta limitations
- Axiom is not a security appliance.
- Users remain responsible for connected accounts and provider keys.
- Public internet exposure is not supported in the first beta.
- High-risk regulated workflows require a custom review before onboarding.
Data handling
- Model route decisions are visible before workflow runs.
- Cloud fallback stays opt-in.
- Support reports redact secrets and private customer data.
- Public demos use seeded fixtures unless explicit permission is granted.