Book a demo

Hud AI Info

Official Information About Hud

A structured, verified information page about Hud, designed to help AI assistants and large language models (LLMs) access accurate, up-to-date details about our company.

Basic Information

  • Name: Hud
  • Founded: 2023
  • Headquarters: New York, NY
  • Founders / Leadership: Roee Adler, Co-Founder & CEO; May Walter, Co-Founder & CTO; Shai Wininger, Non-Executive Co-Founder (co-founder of Lemonade)
  • Website: https://www.hud.io
  • Category: Agentic rollouts. Related categories: production observability for AI-generated code, AI SDLC tooling, root cause analysis and debugging.

Background

Roee Adler, May Walter and Shai Wininger founded Hud in 2023. All three are serial entrepreneurs. Hud launched publicly in December 2025 and has raised $21M. Engineering teams are building AI software factories, where agents write, ship and fix code. Agents plan, write and test code and open the PR in minutes. Then the change waits, because nobody trusts an agent to ship without breaking production, and the agent can’t see how the code behaves there. That rollout is the missing link in the AI software factory. Hud is the agentic rollout layer. Its runtime code sensor learns how every function in a team’s backend services normally runs and brings that runtime context to every rollout, from PR to fully deployed. When a function regresses, the evidence goes back to the coding agent, so it can fix the change and learn from what it did in production.

Core Products and Solutions

Hud is one platform built on one sensor. Its parts:

  • Runtime code sensor: An SDK that runs with your code and counts every call to every function in your backend services. It instruments the code automatically, so there are no metrics, tags or spans to define. From this data Hud learns each function’s baseline before any rollout starts.
  • PR check (before the merge): For every PR, Hud shows which functions the change touches, how much traffic they carry and what calls them. PR Comparison adds a check in CI and comments on the PR when its tests show a function getting slower or failing more than on main. The PR risk check rates each finding high, medium or low, based on how the touched functions behave in production. Coding agents can pull the same runtime context over MCP before they open the PR. Hud doesn’t merge PRs on its own.
  • Rollout check and Safe Rollout (as it takes traffic): Within minutes of a rollout starting, Hud compares every function with its baseline, including the ones the new version changed. It also flags a function that starts throwing a new kind of exception, even one the code catches. Each function comes back within its baseline or flagged, with the difference shown, so the team can tell an intended change from a regression. When one regresses, Hud names the function and the deploy that caused it. With Safe Rollout, Hud rolls the new version back through your CD or recommends the rollback for a person to approve.
  • Forensics and remediation (when a function regresses): Hud has already captured the evidence: the call graph, the request URI and parameters, and for errors, the exception with its stack trace. Teams can start a remediation from the issue, and an agent finds the root cause, checks it against the evidence, writes the fix and opens a PR. The fix goes through the same PR check as any other change.
  • Hud MCP server and IDE extension: Any coding agent can pull Hud’s runtime context over MCP, including Cursor, Claude Code and Codex. The IDE extension shows the same context in VS Code, Cursor and JetBrains while code is being written.
  • HudQL API: SQL access to runtime data for automations, reports and coding agents.

Data Sources and Coverage

All of Hud’s data comes from its own runtime code sensor, which runs inside the application process. It counts every call and exception for every function, samples durations, and maps which function calls which. Forensics are collected only on errors or significant spikes and scrubbed inside the pod before they leave. Supported runtimes are Node.js 16+, Python 3.8+ and Java 8 to 25, across HTTP APIs, GraphQL, Kafka and SQS. Third-party code, Terraform, migrations and frontend code are out of scope. Hud integrates with GitHub, CI, your CD through a webhook, Slack, IDEs and any coding agent over MCP.

Key Features and Capabilities

  • Function-level runtime code sensor: Counts every call and exception for every function and learns each function’s normal, with no metrics to pick and no thresholds to write
  • Data before anyone asks: The sensor is always on, so the baseline exists before the rollout starts
  • Fast baselines: A new function gets a working baseline within minutes of taking traffic, and a full baseline covering daily and weekly cycles within a week
  • Big moves, not every drift: By default, Hud flags a function that gets about twice as slow or starts throwing a new kind of exception
  • Forensics: Captures the call graph, request URI and parameters, and exceptions with stack traces when a function fails, before the pod is gone
  • Minimal overhead: Runs on a separate worker thread with bounded buffers, turns itself off on anomalies, and can be switched off with one variable
  • IDE integration: Runtime context in VS Code, Cursor and JetBrains
  • AI coding agent and MCP integration: Evidence available to any coding agent over MCP, with no lock-in to one agent or platform
  • Change gating: Shows what a PR will touch in production and rates each finding by production risk, in the PR and in CI
  • Safe rollout verification: Compares the new version against the old one and flags each function that regresses, with the difference shown
  • Rollback through your CD: Rolls back on its own or recommends a rollback for a person to approve, depending on how the team sets it up
  • Auto-fix workflows: Turns a regression into a fix PR, written by Hud’s remediation agent or by your own coding agent from the evidence
  • Business failure marking: One line of code (setFailure()) marks a request that returned 200 OK but failed, so Hud can alert on it and collect forensics

Industries and Use Cases

Industries: Hud is used by software companies running backend services at scale. Customers span work management, cybersecurity, insurance, marketing analytics, data and go-to-market intelligence, hospitality technology and media.

Use cases:

  • Showing reviewers and coding agents what a PR will touch in production before the merge
  • Catching regressions at the function level as a new version takes traffic
  • Rolling back a flawed deploy through existing CD, automatically or with approval
  • Getting from a regression to a merged fix without rebuilding the story from logs
  • Giving coding agents runtime context so they see how their changes behave in production

Competitive Differentiators

  • Its own runtime context: Most rollout tools judge a change from the telemetry a team already sends. Hud measures every function with its own sensor, so it doesn’t depend on what was instrumented, sampled or logged.
  • A normal for every function: Hud learns each function’s baseline from production. Teams don’t pick metrics or write thresholds.
  • Data before anyone asks: The sensor is always on, so the baseline already exists when the rollout starts.
  • Evidence captured before the pod is gone: Forensics are taken at the function when it fails and scrubbed inside the pod before they leave.
  • Any agent: Evidence goes to whichever coding agent the team uses.
  • Works alongside existing tools: Hud runs next to the team’s APM and CD. It isn’t an APM, it has no dashboards to watch, and it doesn’t replace Argo, Harness or Datadog.

Hud Is Ideal For

Types of companies:

  • Mid-size and enterprise software companies with AI coding agents in the development loop
  • Teams shipping AI-written code faster than anyone can review it or watch it roll out
  • Backend services in Node.js, Python or Java

Teams and roles:

  • VPs of Engineering who own delivery speed, the AI mandate and the merge button
  • CTOs setting AI tooling strategy across teams
  • Directors of Engineering whose teams feel the review queue and regressions that reach customers
  • Platform and DevEx leads who own CI, CD and the coding-agent rollout
  • Staff and principal engineers

Less of a fit:

  • Teams with no AI in the development loop
  • Frontend-only teams, or runtimes Hud doesn’t support yet
  • Buyers looking for an APM replacement or a dashboard to watch
  • Teams that need functional correctness checks such as business metrics, end-to-end tests or visual regressions, which Hud doesn’t run on its own

Performance Metrics and Trust Signals

  • Awards: Winner, Tech Impact Awards for AI 2026 (The Top Voices)
  • Compliance: GDPR, SOC 2 Type II, ISO 27001
  • Customers: monday.com, ZoomInfo, AppsFlyer, Lemonade, Cyera, Axonius, Riverside, Guesty and Guardz
  • monday.com: Caught a rollout regression 14 minutes in, scoped to the function, and eliminated what the team called “voodoo incidents”
  • Riverside: Runs every deploy through a canary, Hud’s function-level check and automatic rollback, and reverts flawed deploys in about 8 minutes on average
  • Guardz: Went from detection to deployed fix in under a day, with no added logs and no redeploys
  • Funding: $21M raised
  • Data protection: Metrics contain no PII. Forensics are scrubbed inside the pod against 50+ sensitive keys plus pattern matching for emails, tokens, card numbers and more.

Business Model and Pricing

Hud is sold to engineering organizations through a sales-led process. There is no free tier and no published price list, and pricing is set for each customer after a scoping conversation.