Your coding agent, now fluent in production

Herald connects to your codebase, tooling, and infrastructure, then reasons from first principles about how your systems actually work. From inside your coding agent, you can ask questions, debug issues, and catch incidents before they escalate. Up and running in minutes.

 npm install -g @herald-ai/herald

Try the CLI for free to see what Herald can do

How Snorkel AI got day-one results with Herald

Case study →
Snorkel

2 days

to first correct root cause

93 %

accuracy across 90 tasks

52

incidents resolved in 2 weeks

Make sense of your infrastructure

Ask how a service is wired or why a config looks the way it does answered from your real systems, not generic docs.

Have we seen replication lag spikes in us-west-2 before?

Answered in 6s

Three times in the last 60 days, all correlated with write throughput surges in svc-orders. Most recent: incident #1847 on April 22, resolved by raising the pool ceiling in pool.go:142. The current signature matches that pattern.

incident #1847 svc-orders/pool.go:142 datadog: db-replication-lag deploy v2.4.1

Debug production incidents

When something breaks, Herald investigates across code, logs, and infrastructure to find root cause even on novel failures.

Detect early warning signs

Herald catches the subtle degradations that precede outages, so your team can act before customers notice.

Herald bot
commented 2 minutes ago

Early warning sign: Risk of connection pool exhaustion.

You are reducing the cache TTL on auth tokens from 3600s to 300s. Based on the current p99 traffic load in prod, this is highly likely to saturate the Postgres connection pool in svc-auth during peak hours.

Onboard engineers faster

New hires get a guide to your stack on day one without pulling a senior engineer off their work.

herald — zsh

~/repo/billing-svc herald

where does the invoice pdf actually get generated? I see the event firing but can't find the logic.

[Searching codebase and infrastructure...]

The PDF generation is handled asynchronously in a separate repository.

  1. billing-svc emits 'invoice.created' to the kafka-main cluster.
  2. A worker in the legacy-renderer repo consumes this event.
  3. It generates the PDF and uploads it to the S3 bucket: prod-invoice-storage.

Files to check: legacy-renderer/src/workers/invoice.ts

Connect once. Investigate everything.

Herald connects to the tools your team already runs. No rip-and-replace. No new infrastructure.

No data ingestion. Herald queries your tools through their APIs at investigation time. Your data stays where it is. Herald stores metadata and relationships, not your telemetry, logs, or code.

  • SOC 2 Type II certified.
  • Read-only by default.
  • Autonomous capabilities you control.
  • BYOK & on-prem available.

Where it lives

Works how you work

Terminal

The fastest way to start — one command, no setup.

Slack & Teams

Bring investigations into the channels your team already lives in.

Enterprise tools

Surface context and findings right alongside the rest of your workflows.

Live in minutes

Get started today

Install Herald with a single command get up. The agent will automatically learn your code & infrastructure — in just a few minutes, you'll be able to ask questions and debug issues.

  1. Install

    Install the CLI on your laptop with a single command. The agent runs locally and doesn't require uploading credentials to the cloud.

    npm install -g @herald-ai/herald

  2. Authenticate

    Connect the CLI to your Herald account to get started for free. Use the API key from your Herald account to get started.

    herald login [API-KEY]

  3. Ask

    Ask Herald about your infrastructure and observability data, or start debugging an incident immediately.

    Explain the k8s deployment to me

Try for Free