HyperProbe

HyperProbe (Y Combinator S26) attacks the part of an incident that actually costs time: not writing the fix, but finding the value that explains the failure. Most observability tools reason harder over the telemetry you already emit. HyperProbe instead lets your coding agent place a read-only probe on the exact line where the problem occurred in production, and captures the data that was never logged - without redeploying, restarting the service or pausing execution.

How it is used

The SDK installs into a running service, and probes are driven from the editor you already work in. HyperProbe ships an MCP server for Cursor, Claude Code, Codex and opencode, so an agent can set a probe, read the capture and reason about the result inside your existing session. Node.js, TypeScript, Java and Python are supported.

What the probes give you

• Live variable snapshots at a chosen line, including values absent from your logs
• Injected logs and call-stack traces in running production code
• Per-probe configuration for TTL, rate limit and hit expiry, so a probe on a hot path cannot flood anything
• Default PII redaction and an immutable audit log
• 7-day capture history on the free plan, 30 days on paid

The company reports typical time-to-root-cause moving from three or four hours to under ten minutes, with zero redeployments per incident.

Pricing

Priced per service - one running application with the SDK installed, however many instances it runs on - never per engineer or per host. Probes and captures are unlimited on every plan. Free covers one service on managed cloud with full probe configuration, the MCP server, all supported runtimes and community support. Professional is $99 per service per month ($79 billed annually, three-service minimum) and adds unlimited services, 30-day history, team-level probe policy, RBAC, Google and GitHub SSO, a shared Slack channel and self-hosting on request. The first incident is worked with the team for free, and billing is cancellable monthly.

Who it is for

Backend and platform teams carrying a production pager who want the investigation done before a human opens a laptop, and who cannot add a redeploy to every debugging loop.