OpenStatus Releases Free AI On-Call Agent stromboli for Overnight Alerts
A night-time alert usually means the on-call engineer wakes up and scrolls through logs blindly. OpenStatus has released stromboli, a free AI agent that plays the role of an SRE, or site reliability engineer. It kicks in on its own when an alert fires, proposes a quick fix, then digs into logs and metrics for the root cause and posts the findings to Slack. The agent itself is free, but you need your own LLM API key, so you pay for the tokens. Try it first on a non-critical service and work out the token costs.
AI-processed from X: @ChaiWithAi; edited by Hamidun News
Anyone who has been on call for production knows the night alert. A notification fires, the engineer wakes up, opens the dashboards, then the logs, then the dashboards again, and only after about twenty minutes figures out what actually broke. The OpenStatus team proposes handing the first part of this routine to AI. They have released stromboli, an open agent that takes on the role of an on-call SRE. The abbreviation stands for site reliability engineer, the person responsible for keeping a product running and recovering quickly from failures.
What stromboli does during an alert
Stromboli is a self-hosted agent: you run it on your own infrastructure rather than buying it as a cloud service. It triggers when an alert arrives and works in an unusual order. First it suggests a fix that may quickly put out the fire. Only after that does it go looking for the root cause, examining logs and metrics. The agent sends the result to Slack, where the whole team can see it.
The "treat first, understand later" order is close to how working engineers behave. If a service is down, getting it back up matters more to customers than receiving a polished analysis two hours later. Imagine that after a new release the error rate climbs. The agent could propose an option such as a rollback or a restart, then go through the logs and show what changed. This illustrates the logic rather than describing a specific scenario from a real release, but the essence of the work is exactly this: the person in Slack gets a ready hypothesis instead of an empty screen.
Which models and the license
The project is released under the MIT license, meaning it can be freely used and modified. It works with several models: Claude, OpenAI, Gemini, DeepSeek, or local ones. This is convenient for two reasons. First, you are not tied to a single vendor. Second, if log data cannot be sent to external clouds, you can connect a model that runs inside your own perimeter.
There is also an important caveat, which the authors of the review honestly point out. Stromboli is a brand-new project. In addition, it needs your own LLM key, so the agent itself is free, but the calls to the model are paid by you. For a service where alerts fire a couple of times a week, the spending is likely to be modest. But if monitoring is noisy and triggers hundreds of times a day, the token bill is worth estimating in advance.
What to do in practice
If you have on-call rotations and alerts that wake people up, stromboli is worth testing, but cautiously. Here is a sensible plan:
- run the agent on a staging or non-critical service where a mistake costs no money;
- check how closely its first fix suggestions match what a human engineer would do;
- estimate what one alert investigation costs in tokens, and multiply by the real number of triggers;
- decide which model to connect, cloud or local, based on what your logs contain.
Do not give the agent the right to apply fixes to production on its own right away. While the project is new, it is safer to leave the decision to a person: let the agent gather context and propose options, and let the on-call engineer press the button.
The main value of such a tool is cutting the time from "the alert came in" to "it's clear what to do." Even if the agent is wrong half the time, a ready hypothesis and collected logs in Slack save the on-call engineer the first minutes, when their head is not yet working. And open code with support for several models lets you build this into your own processes rather than adapting to someone else's platform.
The bottom line: stromboli will not replace the on-call engineer, but it can become their first assistant. Try it on a small service, calculate the token cost, and only then decide whether to let the agent closer to critical infrastructure.
- Nov 21, 2026FDE job postings grew 4.6x in a year: what a forward deployed engineer actually does at the client's site
- Nov 21, 2026A swimming team captain handed ChatGPT three scheduled tasks and cut her routine work by a multiple
- Nov 21, 2026OpenAI GPT-6 Cyber Found About 35 Flaws in iPhone, Mac and Other Apple Devices
- Nov 21, 202622 Claude Agents Spent Six Weeks on Black Hole Problems — and Solved Two
Key Questions Answered
What stromboli does during an alert?
Stromboli is a self-hosted agent: you run it on your own infrastructure rather than buying it as a cloud service. It triggers when an alert arrives and works in an unusual order. First it suggests a fix that may quickly put out the fire. Only after that does it go looking for the root cause, examining logs and metrics. The agent sends the result to Slack, where the whole team can see it. The "treat first, understand later" order is close to how working engineers behave.
Which models and the license?
The project is released under the MIT license, meaning it can be freely used and modified. It works with several models: Claude, OpenAI, Gemini, DeepSeek, or local ones. This is convenient for two reasons. First, you are not tied to a single vendor. Second, if log data cannot be sent to external clouds, you can connect a model that runs inside your own perimeter. There is also an important caveat, which the authors of the review honestly point out. Stromboli is a brand-new project.
What to do in practice?
If you have on-call rotations and alerts that wake people up, stromboli is worth testing, but cautiously. Here is a sensible plan: run the agent on a staging or non-critical service where a mistake costs no money; check how closely its first fix suggestions match what a human engineer would do; estimate what one alert investigation costs in tokens, and multiply by the real number of triggers; decide which model to connect, cloud or local, based on what your logs contain.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.