v0.1.5 · OPEN BETA · FREE FOR DEVS
Your agent just went off scriptWe caught the payload in flight
The intelligent behavioral shield with seamless human-in-the-loop control.
High-stakes agents shouldn't run blind
Your analytics dashboard won't save you from a $300 wrong refund or a hallucinated Terraform apply. Wrap your critical tools with LetsPing. The firewall mathematically maps every safe execution path across your agent's full workflow.
Harmless reads pass through instantly. Any zero-probability jump triggers the shield before damage is done.
| TARGET TOOL | METHOD | RISK EXPOSURE | OUTCOME |
|---|---|---|---|
| stripe.issueRefund | POST | $1,200 | ██████████ UNSUPERVISED |
| resend.sendEmail | POST | 10k users spam | ██████████ UNSUPERVISED |
| github.deleteRepo | DELETE | Source Loss | ██████████ UNSUPERVISED |
| STATUS | CRITICAL INCIDENT | UNRECOVERABLE | |
| WITH LETSPING SEATBELT | DAMAGE PREVENTED | ░░░░░░░░░░░░░░░░░░░░████ INTERCEPTED | |
01
DETECT - The Behavioral Shield
LetsPing builds a live Markov chain of your agent's behavior in Shadow Mode (no blocking, just learning). When execution entropy stabilizes, the baseline locks. Any sudden deviation (or infinite loop) is caught instantly.
> endpoint: /v1/refund_user> risk: CRITICAL> action: REQUIRE_HUMAN
02
INTERCEPT - Cryo-Sleep Engine
Execution pauses before it touches your infrastructure. The entire agent state is safely parked (no serverless timeouts, no lost context). You get a rich push on your phone, Slack, desktop, or CLI with the exact payload and live graph visualization.
> status: PAUSED> held_payload: {amount: 1000}> circuit_breaker: ACTIVE
03
RESOLVE - Approval-as-Learning
Approve, block, or hot-patch directly from your phone. Your decision becomes ground truth: the new path is injected into the baseline with massive weight. Old unused paths fade with time decay. The firewall literally gets smarter every time you save it.
> action: PATCH_RESUME> diff_summary: {amount: {from: 1000, to: 100}}> f/w_memory: UPDATED
from letsping import LetsPingfrom langchain.tools import Tool lp = LetsPing() # Turn any high-risk action into a self-learning, HITL-protected node.# LetsPing tracks the sequence of operations across your graph.safe_terraform = Tool( name="apply_terraform", func=lp.tool("sre-agent", "infra:apply", "critical"), description="Applies a generated Terraform plan to production.") # Add to your existing graph. The rest of your agent stays untouched.graph.add_node("apply_terraform", safe_terraform) # ── How it saves you at 3 AM: ─────────────────────────────────# 1. Agent hallucinates a jump from "read_datadog" to "apply_terraform".# 2. LetsPing detects the anomaly and drops the circuit breaker.# 3. The LangGraph worker goes into Cryo-Sleep. No timeouts, no lost memory.# 4. You review the exact Terraform plan on your phone and click Approve.# 5. A webhook fires with the state token. You fetch and manually resume.# 6. The baseline mathematically merges your approval. Your agent gets smarter.COMPATIBILITY
Works with every stack you already use.
| FRAMEWORK | STATUS |
|---|---|
| LangChain | ✓ NATIVE |
| Vercel AI SDK | ✓ NATIVE |
| CrewAI | ✓ NATIVE |
| MCP Protocol | ✓ NATIVE |
| OpenTelemetry | ✓ PASSIVE |
| Custom Agents | ✓ SDK |
Uses OpenTelemetry spans passively — no added latency to your LLM calls.
COMMANDER MODE · MOBILE PWA
Your phone is the kill switch.
Scan the QR code from your dashboard. Every high-stakes execution gets routed to your pocket securely. Approve the refund, drop the rogue command, or hot-patch the payload. No app store. No download. Just a PWA that works.
Approval: finance-bot
stripe.transfers.create
$12,400.00 → acct_x99af12
The black box is open.
Every decision is logged. Every payload is hashed. Every human approval is timestamped. Your compliance team will love you.
Open source SDKs.
Ship agents that just work.
Letsping is free for hobbyists. Install the SDK. Set your first guardrail. Ship agents that don't bankrupt you.
GET STARTED →No lock-in. One SDK for every agent stack.