We’re excited to share that RunLLM is now Herald.
Every blog post about an AI company starts with a meditation on an era of unprecedented change. We’re sick of it, and you don’t need to be told again. We’re more interested (and impatient for) what comes next.
The interesting question isn’t whether AI generated code is going into production (it is), or whether it will cause more breakage (it will), or even how we’re going to handle those breakages (agents). This is all obvious to you, to us, and to anyone who has been thinking about software engineering for the last 12 months. Those changes are already here.
Once that’s clear, the next obvious question is not whether you use an agent but what that agent actually does. That’s what we’re interested in. The rest of the world has set out to build agents that blindly recreate all the frustrations and shortcomings in existing observability process – alert fatigue, outdated runbooks, customer-found defects, and absurd storage and indexing costs.
That’s the wrong way to approach the problem. Agents don’t just change your ability to respond to your existing alerts – they can continuously evaluate your systems and tell you what’s going wrong before anyone else does. That’s what we’re building with Herald.
Why Herald
herald | ˈherəld |
noun
1 an official messenger bringing news
verb [with object]
be a sign that (something) is about to happen: the speech heralded a change in policy.
Traditionally, a herald announced news. And colloquially, the word herald indicates something is about to happen. Both suggest proactivity. You're not waiting for something to happen and then reacting to it; you're indicating what's about to. That perfectly describes where observability is heading.
What we do
Herald is built with three goals in mind. First, improve reliability. Software is changing faster than ever but reliability is still critical. Second, reduce toil. Engineers will need to be involved in making reliable software but certainly don't need to be scrolling through hundreds of dashboards or writing custom log queries. Third, share knowledge. The best engineering teams are fully up to speed on what's going on, but writing and sharing is hard.
Our brand reflects how our product is built, performs in real production environments, and how we’ll continue to bring value to our customers.
Herald is proactive in two key ways:
- It doesn’t wait to be told what to do, it learns. The standard approach in AI SRE agents has been to write runbooks and have an agent execute the contents of a runbook. This works great if your engineering team loves writing documentation – in other words, it doesn’t work. Herald doesn’t need to be told what to do: It learns on its own by cataloging and understanding the relationships between different parts of your product, infrastructure, and business. That understanding enables it to formulate hypotheses and investigate effectively (70%+ accuracy on novel incidents) without runbooks.
- It doesn’t wait for alerts, it detects. Observability by its nature has also historically been reactive. Software systems generate too much data for humans to process, so you summarize that data, set thresholds that indicate “bad enough,” and wait for alerts. That means that you’re usually finding out well after the fact that something has broken. Herald doesn’t wait: It uses statistical anomaly detection to find early warning signs, validates those warning signs with other system signals, investigates the root cause, and surfaces incidents to you before they get worse or customers complain.
Seeing is Believing
The problem with AI SRE today is that it's all fluff with little substance. Most agents take weeks or months to just get set up, POCs drag on for multiple quarters, and everything requires perfect documentation. And this doesn’t even address the realities of going through legal, security or procurement processes just to see if something might work on your stack.
That’s too slow, and it’s not how engineering teams work today. We decided to fix that problem.
With the Herald CLI, you can get up and running almost instantly. If you have 5 minutes, you can install the agent, configure it, and be able to answer questions about your infrastructure, debug incidents, and even get proactive notifications when things are breaking. For free.
The Herald CLI runs on your laptop and doesn’t upload credentials to the cloud.
👉 To join the waitlist for Herald CLI, sign up here.
Once cleared to use the product, all you’ll have to do is run npm install @herald-ai/herald, and start the CLI. The agent will learn how your codebase is structured, what infrastructure and observability you have, and how it should connect to your tools.
Happy Heralding!