HyperProbe – AI On-Call Agent for Production Debugging

TL;DR

TL;DR: HyperProbe embeds an AI-native debugging agent into your production runtime — probes capture live variables and stack traces without pausing execution or requiring a redeploy.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: hyperprobe.co
• Pricing page: hyperprobe.co/pricing
• Docs: docs.hyperprobe.co
• HN launch thread: news.ycombinator.com/item?id=49185389
• YC batch: S26 (verified via HN thread title)

What Is HyperProbe?

HyperProbe is an AI on-call agent for production debugging. Instead of adding print statements, restarting services, or opening support tickets, you drop an SDK into your runtime and let an AI coding agent — Cursor, Claude Code, Codex, or any MCP-compatible agent — inspect live variables, trace call stacks, and inject logs on demand.

The core guarantee is read-only, non-disruptive debugging: probes capture what they need and clear themselves automatically. No side effects, no latency spikes beyond a bounded overhead cap, no redeploys.

The product was founded by the team behind ShailendraHT and backed by Y Combinator (S26 batch).

Setup Workflow

Step 1: Install the VS Code Extension

Search for HyperProbe in the VS Code Marketplace and install it. Alternatively, use Quick Open (Ctrl+P / Cmd+P) and run:

ext install hyperprobe.hyperprobe-extension

Click the HyperProbe debug icon in the Activity Bar, then click Login. You’ll be redirected to the dashboard at app.hyperprobe.co to sign in via SSO and create a service.

Step 2: Configure Your Service

Copy your Service UUID from the dashboard. Create a .hprc file at the repository root:

{
  "serviceId": "<service-uuid-from-dashboard>"
}

Step 3: Add the SDK Dependency

Node.js:

npm install @hyperprobe/node-sdk
# or
pnpm add @hyperprobe/node-sdk
# or
yarn add @hyperprobe/node-sdk

Java: Download the agent JAR from your dashboard:

curl -O https://hyperprobe-assets.s3.ap-south-1.amazonaws.com/static/hyperprobe-agent.jar

Step 4: Initialize the SDK

Node.js (TypeScript) — create src/hyperprobe.ts and import it early in your main entrypoint:

import { HyperProbe } from '@hyperprobe/node-sdk';

HyperProbe.start({
  serviceId: '<service-uuid-from-dashboard>',
  environment: process.env.NODE_ENV,
  brokerUrl: 'https://logger.app.hyperprobe.co',
  commitSha: process.env.GIT_COMMIT,
  distLocation: 'dist',
  sourceMapDir: './dist'
});

Java (JVM): Initialize the agent in your launch routine per the docs.

Step 5: Connect Your AI Agent

HyperProbe ships an MCP server that any MCP-compatible AI coding agent can connect to directly. The agent gets live access to probe configuration, capture history, and variable inspection — without leaving the terminal or editor.

Pricing

Priced per service (one running application, however many instances):

| Plan | Price | Services | Capture history | |------|-------|----------|-----------------| | Free | $0 forever | 1 | 7 days | | Professional | $99/service/month (annual: $79) | Unlimited | 30 days | | Enterprise | Custom annual | Unlimited | Custom |

All plans include unlimited probes and captures — nothing about debugging is metered. Probe bounds (TTL, rate limit, hit expiry) are set by the operator, not the vendor.

How Probes Are Controlled

Every probe is bounded by controls you define:

• TTL — probe clears itself after N seconds/minutes, whether or not it fired
• Rate limit — caps how often a probe captures, keeping high-traffic lines bounded
• Hit expiry — probe clears itself after N captures, auto-cleanup after data is collected

Enterprise plans add org-level policy ceilings: your security team sets maximum TTL and mandatory expiry at the organization level so no developer can place a probe outside policy.

Practical Evaluation Checklist

• [ ] VS Code Extension installs and connects to dashboard
• [ ] .hprc maps local workspace to service UUID correctly
• [ ] Node.js SDK initializes without restart
• [ ] Probe placed and captures variable snapshot within TTL
• [ ] Probe clears itself after TTL / hit limit
• [ ] MCP server connects in Cursor / Claude Code / Codex
• [ ] Capture history persists across team members (Professional)
• [ ] PII redaction rules applied correctly

Security Notes

• Read-only by default — probes never write to your application state
• Immutable audit log — all probe placements are logged
• PII redaction — configurable rules redact sensitive fields before capture
• RBAC — Professional and Enterprise support role-based access control and SSO (Google, GitHub, SAML)
• Self-hosted / private VPC — Enterprise plan supports on-premises deployment

FAQ

Q: Does HyperProbe introduce latency to my production service? A: HyperProbe is designed for minimal overhead. The agent activates a cooldown shield if overhead exceeds 15ms or free memory drops below 15%, preventing the probe from interfering with production performance.

Q: Can I use HyperProbe with any runtime? A: Node.js and JVM (Java agent) are officially supported. The MCP server is runtime-agnostic and works with any MCP-compatible AI agent including Cursor, Claude Code, Codex, and opencode.

Q: Who can place probes? A: On Free and Professional plans, any team member with dashboard access can place probes. Enterprise plans add org-level policy ceilings so security teams can enforce maximum probe bounds organization-wide.

Q: Is there a self-hosted option? A: Yes, Enterprise plans support self-hosted deployment or private VPC. Contact HyperProbe for details.

Conclusion

HyperProbe closes the loop between AI coding agents and production debugging. Instead of guessing from logs or adding instrumentation that requires a redeploy, your agent can directly inspect live state through bounded, self-clearing probes. The Free tier is genuinely useful for single-service hobby projects; the Professional tier unlocks team collaboration and longer capture history at $99/service/month.

If you run production services and your team uses AI coding agents, HyperProbe is worth integrating into your on-call workflow.

HN thread: news.ycombinator.com/item?id=49185389