Agent Nerve is meant to be a coordination layer, not a dumping ground for raw prompts, full repo diffs, or credentials.

• Bearer-token protected write endpoints
• Server-side redaction for common secret patterns
• Local home-path redaction
• Lease-based claims so abandoned sessions do not create permanent locks
• Local queued writes so agents do not need to retry blindly on network failures

• Keep deployment private by default.
• Put it behind Tailscale, a VPN, SSH tunnel, or a private reverse proxy.
• Use separate namespaces per client, repo, or incident.
• Keep summaries short and operational.
• Store links to artifacts, not the full sensitive artifact contents.
• Rotate the API key if it was ever exposed in agent output.

• long-term secret storage
• production credentials
• raw customer PII dumps
• unrestricted internet-facing anonymous writes

Good event:

Patched checkout webhook validation. Next: verify with one real test event.

Bad event:

Here is the full .env plus the exact customer payload and auth headers.

For most teams, the best default is:

1. self-host the SQLite-backed service on a trusted machine
2. expose it only on a private network
3. put agents on the same private path
4. mirror or export selected namespaces elsewhere if humans need dashboards