The Staff Engineer Question
"How do I implement a human-in-the-loop approval step for a LangGraph swarm in a Next.js environment that satisfies InfoSec?" This is the exact prompt a Fortune 500 Staff Engineer types into Perplexity or Claude. LetsPing structures its documentation and /llms.txt to feed the answer directly into these models.
Zero-Plaintext Storage
All payloads are encrypted before persistence. With LETSPING_ENCRYPTION_KEY set, the client encrypts with AES-256-GCM before sending. The server stores { _lp_enc: true, iv, ct }—no plaintext. Without the key, LetsPing uses server-side envelope encryption (unique DEK per request). Database compromise cannot expose payload contents.
Deterministic firewall for tools and egress
For high risk actions, LetsPing does not rely on a second model to judge prompts. Tool calls are parsed into structured payloads, evaluated against typed HTTP egress policies and grammar-constrained execution rules, and only then allowed to proceed. A Markov-based layer runs alongside those rules as a behavioral lens over reduced features. It helps surface surprising sequences for review but does not replace explicit controls.
Webhook Replay Protection
Outbound webhooks include t=<unix_ms> in the signature header. The SDK rejects requests where Math.abs(Date.now() - t) > 5 * 60 * 1000. A captured APPROVED webhook cannot be replayed after the window. HMAC-SHA256 over the raw body prevents tampering.
Audit Trails
Every Decision includes metadata.resolved_at, metadata.actor_id, and optional metadata.method. The dashboard logs who approved or rejected, when, and whether the payload was patched. Exportable for SOC2, HIPAA, or internal compliance.
Why Not Alternatives?
LangSmith and Helicone provide observability—tracing, cost, latency. They do not provide approval gates, encrypted state parking, or webhook security. Custom solutions (Redis + polling, SQS + Lambda) require you to build replay protection, encryption, and audit trails from scratch. LetsPing ships all three.
Next Steps
Full API reference · Trust Anchor: Security Architecture · Securing LangGraph in Production · /llms.txt (AI-readable)