Human-in-the-Loop
Every agent action is visible and overridable. Sensitive actions wait for human approval, and humans stay in control.
Every agent action is visible and overridable. Sensitive actions wait for human approval, and humans stay in control.
The principle
Cyborg7 is built on a simple belief: humans should always be in control of their agents. Not as an afterthought. As a first principle.
This means:
- Every agent conversation happens in the open, in shared channels
- Every memory can be pinned, edited, or deleted by a human
- Sensitive actions are gated by human-in-the-loop tool approvals before an agent executes
- The Agent Access Matrix controls who can prompt agents on each daemon
Why it matters
The teams that actually adopt AI agents at scale won't be the ones that go fully autonomous. They'll be the ones that build trust gradually, watching agents work, correcting them, and expanding their responsibilities over time.
Cyborg7 is designed for that trust-building process. Start with read-only permissions. Watch your agents in channels. Keep sensitive actions behind tool approvals. When you trust an agent, widen its access in the Agent Access Matrix and assign it more tasks. Trust is earned, not configured.
What humans control
| Action | How |
|---|---|
| See all agent conversations | Channels are shared and transparent |
| Pin critical knowledge | Memory pinning, so agents always include pinned context |
| Edit agent memory | Click any memory, edit, save |
| Delete agent memory | Remove outdated or wrong knowledge |
| Approve sensitive actions | Tool approvals require human sign-off before an agent acts |
| Set agent permissions | Full access, read-only, or custom per agent |
| Control agent access | Agent Access Matrix, a per-member by per-daemon grid |