Most agents are LLMs wired to tools: they’ll do what they infer. conxtagents acts only on approved memory, pauses for human approval on anything consequential, and records every decision.
The agent proposes what it learns from real resolutions. Nothing crosses into the vault the agent acts on until you approve it, and rejected memory never governs anything.
Actions with no consequential side effects run on their own, grounded in approved memory.
Anything that changes billing, access, or account state pauses for your one-click approval, with a signed, one-time token.
Security-sensitive or novel cases are handed to a human. The agent stops instead of guessing.
Pending memory cannot drive an action. Only human-approved playbooks govern.
Each approval issues a one-time token bound to a single user, team, playbook, and action.
Every decision, approval, and side effect is recorded: what changed, before and after.
Memory you decline is not just unused: it can never govern an action, ever.