Trust & Security
Autonomy you can audit.
Volobox treats agents like employees: they have an identity, permissions, a manager, and limits. Trust is engineered into the platform, not promised in a policy document.
The model
Three modes of agent authorization
Pick per agent, per situation: from fully supervised to independent within explicit boundaries.
User-delegated
The agent acts strictly as an extension of the person who asked, with their permissions. Every write produces an approval card that person must accept. The default, and mandatory for personal assistants.
Manager-escalated
The requester's permissions are tried first. When they are not enough, the agent's designated manager approves, or a standing rule the manager owns pre-approves it within a budget.
Self-authorized
The agent is a first-class principal with its own memberships in projects, spaces, and drives. Inside that boundary it works without approval cards (like an employee) under mandatory rate limits.
Approvals
Every write asks, until you decide it doesn't have to
In Agent mode, write actions become proposal cards: see exactly what will happen, approve or reject, and get a deep link to what was created. Ask mode is fully read-only: write tools are simply not available.
- Standing auto-approval rules with per-hour, per-day, or lifetime budgets
- Manager escalations and agent questions arrive in your Inbox
- Agent-wide safety limits cap writes per run and per hour (hard-enforced)
Transparency
See everything, after and during
Live execution
Watch coding sessions in real time: terminal, diffs, plans, questions, per-turn cost.
Run traces
Every agent run keeps a trace of tool, MCP, and sandbox calls.
Audit log
Authorization decisions and security events, exportable as CSV/JSONL or pulled into your SIEM over the API.
Cost visibility
Usage and cost per model and per conversation in admin settings.
Activity history
Agent actions appear in the same feeds as human actions, attributed to the agent.
Personal-data gates
Shared agents cannot reach personal data; separate gates guard personal tools.
Isolation & data control
Your data stays yours
Sandbox isolation
Agent code runs in isolated sandboxes with budgets, idle pause, and destroy controls, in the cloud or on machines you own.
Machines you own
A personal machine runs only its owner’s private work. Its runner connects out, opens no port, and accepts only updates signed by Volosoft.
Outside people stay outside
Customers reach only their Portal. Every workspace address is refused for them, and internal notes never reach it.
Memory that respects access
Personal memory is private to its owner, and a shared note only reaches people who could open what it came from.
Scoped secrets
Encrypted secret scopes; tokens never travel inside packs or exports.
Deployment freedom
Cloud or on-premise. Bring Claude, GPT, Gemini, Kimi, or an open-source model you host. Workspace knowledge indexing runs inside your deployment.
Calendar privacy
Teammates see a synced calendar event only as Busy and a time: never its title, place, guests, or description.
Sessions that end
Signing out ends the session on the server too, and sign-in sessions no credential can still use are removed.
Spend limits
Usage limits warn administrators and stop new work when AI spend, sandbox compute, or workflow runtime reaches its cap.
Enterprise security
For your security review
Engineered in today: role- and permission-based access control, OpenID Connect authentication, Microsoft Entra ID sign-in, an exportable audit log with API access for your SIEM, encrypted secrets, isolated execution, and on-premise deployment.
Evaluating Volobox? Talk to us about sign-in options, data residency, network isolation, and our compliance roadmap. We'll walk through the architecture with your security team.
Bring your security team to the demo.
The governance model is best shown live: approval cards, limits, and audit included.