# HyperProbe

> AI-native production debugger that places read-only virtual breakpoints on live running code to capture exact variable state at the moment of failure—no redeployment or restart required.

HyperProbe is a Y Combinator-backed AI-native production debugger that eliminates the log-redeploy-wait cycle engineers face during incidents. It works by dropping read-only virtual probes on exact lines in live running services, capturing variable state that logs never record—without restarting or redeploying the service. The tool supports JavaScript, TypeScript, Java, Kotlin, Python, and Ruby, and integrates with coding agents like Cursor, Claude Code, Codex, and Opencode.

## What It Is

HyperProbe sits in the production debugging category, specifically targeting the gap between what observability tools (logs, traces, metrics) surface and what engineers actually need to confirm root cause. Traditional debugging requires adding log statements, pushing a new build through CI/CD, and waiting for the failure condition to reproduce—a cycle that can take two to three hours per incident. HyperProbe replaces that cycle with a "probe": a non-blocking, read-only snapshot of live variable state at a specific line, triggered by real traffic. The probe fires asynchronously, captures the exact values at that moment, and then disappears. According to the vendor, overhead is under 1% at 3,000 RPS on their reference service.

## How the Probe Workflow Works

When an alert fires from PagerDuty, Datadog, or Slack, HyperProbe's agent reads distributed logs and traces to locate the suspect file and line. It then places a virtual breakpoint at that location—no code change, no redeploy. When real traffic hits that line, the probe captures the live variable state and delivers an evidence-backed root cause analysis. The vendor describes the end-to-end flow in five steps: Alert → Plan → Probe → Capture → Confirm. In their published example walkthrough, the time from alert to confirmed root cause is under ten minutes.

## Architecture and Security Model

HyperProbe is designed around a read-only constraint enforced at the architecture level:

- **Read-only by design:** The agent can capture state but cannot write memory or execute code.
- **Immutable audit log:** Every probe placement and capture is logged and cannot be altered.
- **Approval gates:** On enterprise deployments, probes must be approved before they arm.
- **PII redaction:** Default redaction runs at the agent before capture on every plan; enterprise adds custom rules per field and per file.
- **Self-hosted or private VPC:** Available on enterprise, with data residency in a chosen region and no captured values leaving the customer's network.
- **Probe controls:** Every probe carries a TTL, rate limit, and hit expiry that auto-clear the probe when reached.

## Failure Types It Targets

HyperProbe is specifically designed for failure modes that traditional observability misses:

- **Silent failures:** Returns 200 with wrong body; trace is green; value was never logged.
- **Exceptions far from cause:** Stack trace points to line 82 but the cause is at line 18 or in a different file.
- **Swallowed exceptions:** Exception caught internally, no alert, no error, only a business metric movement.
- **Race conditions and duplicate processing:** Requires thread state at the exact moment of overlap.
- **Third-party contract drift:** Vendor added a new field or status value with no handler in the code.
- **Business metric drops:** Payments failing or orders dropping with no exception anywhere in the stack.

The vendor also lists memory leak diagnosis, OOM root cause, CPU spike isolation, and latency spike tracing as capabilities shipping in the near term.

## Coding Agent Integration

HyperProbe exposes an MCP (Model Context Protocol) server that coding agents use to drive probes. Installation is designed to be agent-driven: pasting one prompt into a coding agent pointing at the installation guide causes the agent to install the SDK, wire the entrypoint, enable source maps, and register the MCP server. The vendor states most teams are capturing within fifteen minutes. The MCP server is included on every plan, including the free tier, because—as the pricing page notes—"gating it would leave the free plan with nothing to do."

## Current Status

HyperProbe is actively available with a self-serve free tier and a paid professional tier. The product is backed by Y Combinator. Go runtime support is listed as on the roadmap. The vendor offers a proof-of-concept engagement where they work one real incident from a customer's stack in a 30-minute call, with a confirmed root cause delivered before the call ends or no further discussion required.

## Features
- Read-only virtual breakpoints on live production code
- No redeployment or service restart required
- AI agent-driven probe placement via MCP server
- Automatic alert ingestion from PagerDuty, Datadog, and Slack
- Distributed trace analysis to locate suspect lines
- Live variable state capture at exact moment of failure
- Immutable audit log of all probe activity
- PII redaction at the agent before capture
- Approval gates before probe arms (Enterprise)
- RBAC and SSO support
- Self-hosted or private VPC deployment (Enterprise)
- Probe TTL, rate limit, and hit expiry controls
- Unlimited probes and captures on all plans
- Supports JavaScript, TypeScript, Java, Kotlin, Python, Ruby
- Works with Cursor, Claude Code, Codex, Opencode
- Under 1% overhead at 3,000 RPS
- Shared workspace and saved probes (Professional)
- Custom PII redaction rules per field and file (Enterprise)
- Data residency in chosen region (Enterprise)
- SAML SSO and SCIM provisioning (Enterprise)

## Integrations
PagerDuty, Datadog, Slack, Cursor, Claude Code, Codex, Opencode, GitHub SSO, Google SSO, SAML SSO, SCIM

## Platforms
WEB, API, CLI, DEVELOPER_SDK

## Pricing
Freemium — Free tier available with paid upgrades

## Links
- Website: https://hyperprobe.co
- Documentation: https://docs.hyperprobe.co/quickstart
- EveryDev.ai: https://www.everydev.ai/tools/hyperprobe