Docs
Start typing to search…

FAQ

Hud helps engineers understand how their code behaves in production - at the function level. This FAQ covers common questions about the integration and product.

⚙️ How Hud Works

What is Hud?

Hud is the first Runtime Code Sensor designed specifically to make AI-generated code production-safe by default. It helps engineers and Agentic AI understand how their code behaves in reality - in production. It automatically collects function-level performance and error data, builds a call graph, and correlates performance with changes in usage and code.

How does Hud work?

Hud uses a lightweight language-specific SDK that auto-instruments your code (without modifying it), collects invocation counts, durations, exceptions, and more - and sends that data to Hud’s backend. It builds an aggregated, always-on function-level call graph of your system.

What does Hud measure?

Hud captures for each endpoint, queue and function:

  • Invocation counts
  • Execution durations
  • Exceptions (at line level)
  • Call relationships between functions
  • Correlation with code changes, request types, and usage patterns

On error, Hud automatically collects context needed to solve it including:

  • Hud functions' invocations call graph
  • HTTP URI
  • HTTP parameters
  • Exception-throwing functions
  • Thrown exceptions: error type, stacktrace and error message

Does Hud use tracing?

No. Hud does not use distributed tracing. Instead, it collects function-level telemetry and builds an aggregated call graph over time - with very low overhead (1-2%).

What kinds of systems does Hud work in?

Hud works in:

  • HTTP APIs (REST, GraphQL, etc.)
  • Async workers (Kafka consumers, SQS consumers, etc.)
  • Any user-defined functions

It ignores third-party code (node_modules, site-packages, etc.).

What’s the overhead of using Hud?

Hud adds about 1-2% CPU overhead. It is designed to be safe to run in production at all times.

Is sampling involved?

Hud captures all invocation counts and all exceptions.
For durations, Hud uses incremental sampling - recording timings on an increasing fraction of invocations (e.g. earlier calls are measured more frequently, later calls less frequently). This ensures representative performance data with minimal overhead.

Where is the data shown?

You can access your data:

  • In Hud’s web app
  • In your IDE (via the Hud extension) - right in the code editor
  • In your Agentic AI - using Hud's MCP

🚀 Installation & Setup

Which runtimes are supported?

Hud currently supports:

  • Node.js (>= 16.x)
  • Python (>= 3.8.x)
  • Java (coming soon)
  • C# (coming soon)

More language support is planned.

Which IDEs are supported?

Hud provides an IDE extension for:

  • VSCode
  • Cursor
  • JetBrains IDEs (e.g. WebStorm, PyCharm, IntelliJ IDEA, etc.)

How do I install the Hud SDK?

Node.js SDK Installation
Python SDK Installation

Is sourcemap support included?

Yes — Hud supports source map uploads so that you can see accurate line-level data even for minified or transpiled code.

🔒 Safety & Compatibility

If Hud’s backend is down, will my app be affected?

No — the Hud SDK is designed to fail-safe:

  • If Hud’s backend is unavailable, your app will continue running normally
  • Telemetry is buffered in memory (not persisted to disk) and discarded if needed - no blocking

What safety measures does Hud’s SDK implement?

  • Data is sent from a separate worker thread to isolate memory, CPU, and event loop usage.
  • You can disable instrumentation via the HUD_ENABLE environment variable, ensuring opt-out without code changes.
  • Hud self-monitors and can auto-disable if anomalies are detected.
  • The SDK is under continuous internal monitoring (memory, CPU, ELU, etc.).
  • Graceful shutdown (app runs even if backend is unavailable).
  • Telemetry buffers are memory-safe and bounded.
  • Timeouts and rate limits to ensure telemetry does not impact app performance.
  • Minimal overhead (1-2%).

I already have an APM tool — will that be a problem?

No — Hud is designed to work alongside existing APM tools.
It does not interfere with other observability agents or tracers - and focuses on providing deeper function-level insights, which most APM tools do not.

How will I know about maintenance or downtime?

Hud actively monitors all customer pipelines.
If data flow is interrupted, you’ll receive detailed updates about the impact (e.g., data loss or lag) and resolution steps.

Is the SDK safe to run in production?

Yes - Hud's SDK runs on millions of services every day, and is built to run in production with minimal overhead and safe error handling.

What about privacy and security?

Hud only collects runtime metadata — function names, files, durations, and errors — and automatically scrubs any sensitive data before sending.

Hud complies with GDPR, SOC2 type II, and ISO 27001 and offers strong security practices (ask us for our security whitepaper).

Updated 1 day ago

What’s Next