Best answer
Use Nerve when an AI or automation runner should notify a human about failure, approval required, risky output or completed remediation. Nerve is not the AI agent. It is the encrypted phone delivery layer and optional signed-action boundary around automation.
Good AI-agent signals
- A coding agent opened a pull request but tests failed.
- A maintenance agent detected a production issue and stopped before applying a fix.
- An automated deploy agent finished rollback and needs human review.
- A scheduled repair job exceeded its budget or hit an unknown state.
- An E2E or smoke-test agent found a user-visible failure after deploy.
Minimal pattern
export NERVE_DSN="nerve://TOKEN:[email protected]"
./agent-task.sh || printf 'agent task failed\nrun: %s\n' "$RUN_URL" \
| nerve send --severity critical --title "Agent task failed"Boundary
Give the agent runner a sender DSN, not a broad chat bot token, read token, SSH key or agent token. The sender DSN can only send encrypted signals into one pipe. If a trusted host should run bounded actions later, use the separate Nerve agent path and signed command payloads.
Do not use Nerve as
- The AI reasoning engine.
- A queue for long agent transcripts.
- A replacement for CI logs, test artifacts or code review.
- An unrestricted remote terminal from a phone.