What Livops's AI assistant does when a build fails
When a build fails on Livops, the previous version keeps running and one button asks the assistant why. It reads the failed build's log and your project's commands, says what went wrong and how to fix it, and may offer one fix as a button that changes nothing until you press it. It never deletes, never sees variable values, and does not edit your code.
By Livops team, the people who build and run LivopsUpdated
A failed build is where an AI assistant is most useful and where it is most tempting to oversell one. This page describes exactly what the assistant built into the Livops cabinet does when your build fails: what it reads, what it says, what it can offer to do, what it never does, and where the line sits between a suggestion and an action.
What happens when a build fails?
Before the assistant does anything:
- The previous version keeps running. A failed build, a start that does not answer, or a build past ten minutes leaves the live version where it was.
- The cabinet says why. At the top is an error code and a short reason, with the most useful line of the log under it. Many failures end the log with a Tip: line that says what to do. The Details link opens the page for that code in the deploy errors guide.
- The full log is there. The end of it, the last 10 to 20 lines, is usually where the error is.
Next to a failed build there is one more button: ask the assistant.
What does the assistant read?
When you press it, the assistant is given:
- the failed build’s own log, as plain text;
- the project’s type and template;
- its install, build, start and pre-deploy commands;
- the error code the build stopped with.
It is handed variable names, never their values. It only explains a build that failed: asking a model to find a problem in a successful log is asking it to invent one, so a build that worked is refused.
What does it say?
What happened and how to fix it, in plain words. The answer is saved with the build, so pressing the button again is free and returns the same answer. Each account can ask for up to 20 new explanations a day.
When the cause is in your code, a missing package, a type error, a command your project does not have, the assistant tells you what to change and where. It does not change it: none of its buttons writes to your repository, so the fix goes through your editor and a push, like any other change.
Can it fix the build?
It can offer one fix as a button, and most of the time it offers none. After a failed build, the button can only be one of these:
| Button | Offered when the log shows |
|---|---|
| Retry the build | Something that could go differently a second time: out of memory, a failed download, a registry or network timeout. Never for a mistake in the code |
| Set the project folder | The repository has nothing at the top and names the folders that do, such as backend and frontend |
| Set the output folder | The build worked but wrote its files somewhere the settings do not name |
| Change the commands | The build, start, install or pre-deploy command itself is wrong and the right one is visible |
| Add a database | The app needs a database it was never given, such as a refused connection to port 5432 |
| A larger database disk | The database ran out of disk. The card shows the new monthly price, and only the team owner can press it |
| More memory | The machine killed the process for memory |
When two answers are equally likely, two folders for example, the assistant is told to name both in words and offer no button: guessing would put the wrong site online, and that looks like success. A button that is not about the build you are looking at is dropped before you see it. A saved answer’s button is checked again against today when you reopen it, so a database resized since then does not get offered twice.
Some failures are already recognised by the platform itself, such as a process killed for memory. When the assistant explains one of those and offers nothing, Livops adds the button that fits, checked the same way.
What happens when you press the button?
A regular cabinet action runs. Your session and your permissions are checked again, exactly as if you had found that setting and changed it yourself. The button carries no permission of its own, so an explanation cannot make the cabinet do something you are not allowed to do.
The chat in the corner of every screen can talk about a failed build too: when the conversation is about one, the chat is given that build’s last lines, and a fix it proposes is a card you press. One chat card behaves differently, and it is not about failed builds: when you ask the chat to deploy one of your own repositories or a folder you attached, the new project is created and its first build starts as soon as the answer arrives, because that is what you asked for. After a failed build, a card that would create a new project is never offered.
Is it self-healing?
No, and we do not call it that. By default, nothing the explanation offers runs until you press its button.
There is one setting that changes this, and it is off unless your team turns it on: auto mode, under the team’s Settings → Assistant. With it on, seven small, reversible actions run without a press: retrying or cancelling a build, restarting, adding a domain, rechecking DNS, and setting the project folder or the build output folder. Everything that costs money or can take a site offline still waits for you, including deploys, rollbacks, stopping a service, changing commands, adding or growing a database or a storage bucket, variables and promoting staging.
What does it never do?
- It never deletes anything and never restores from a backup.
- It never sees variable values, only their names.
- It does not write a database address or password into a static site.
- It does not edit your repository.
Privacy and limits
When the assistant is on and you ask it something, your question, your project names and the failed build’s log are sent to the AI provider, Anthropic. To keep that from happening, turn the assistant off under Settings → Assistant; after that the server sends nothing. Failed build explanations are limited to 20 a day per account. Chat questions have a monthly allowance per plan, listed in the assistant docs.
From your own AI agent
If you work in Claude Code, Cursor, Codex or VS Code, the same explanation is one tool call away. Livops’s MCP server has explain_failure, which asks the assistant why a failed deploy failed and returns the answer to your agent; it needs the team’s assistant on, and it is rate limited. Anything your agent does next, such as changing code and pushing or calling deploy, is its own tool call, and your client can ask you before it runs.
Seeing it on Livops
- Deploy a project from GitHub, an uploaded folder or a template.
- If a build fails, open it: read the reason, the highlighted line and any Tip:.
- Press the assistant’s button next to the failed build, read the explanation, and press the fix if one is offered and you agree with it.
More in AI assistant and Deploy errors. Connecting your own AI client is described in Deploy from Claude Code.
Questions
Does the assistant fix failed builds automatically?
No. It explains the failure and may offer one fix as a button; the fix runs when you press it. The one exception is auto mode, which is off by default: a team that turns it on lets seven small, reversible actions run on their own, such as retrying a build or setting the project folder. Anything that costs money or takes a site offline still waits for the button.
Does it change my code?
No. None of its buttons writes to your repository. A mistake in the code, such as a missing package or a type error, is explained in words for you to fix and push. Its buttons change settings, such as the build and start commands or the folder the project lives in, or run something again.
What does it read?
The failed build's log, the project's type and template, its install, build, start and pre-deploy commands, and the error code the build stopped with. It sees variable names, never their values.
Is my build log sent to an AI provider?
Yes, when the assistant is on and you ask: your question, your project names and the failed build's log go to Anthropic. Turn it off under the team's Settings, Assistant, and the server sends nothing.
How many explanations can I ask for?
Up to 20 a day per account. An explanation is kept, so pressing the button again on the same build is free and gives the same answer.
Can my AI coding agent get the same explanation?
Yes. Livops's MCP server has an explain_failure tool that returns the assistant's explanation of a failed deploy to Claude Code, Cursor, Codex or another client. It needs the team's assistant on, and it is rate limited.
Does it ever delete anything or restore a backup?
Never. There is no delete among its actions and no restore from a backup, by design: one wrong answer must not cost anybody their data.
Put your project online today
Connect GitHub or drop a folder. No card is asked for.
Deploy and see itRelated
- MCP server for deploying apps: what to look forAn MCP deploy server lets Claude Code, Cursor, Codex or Copilot deploy, read logs and roll back on your hosting account. Judge one by how it signs in, how far a key reaches, whether its tools say which ones change things, and what you can see and revoke afterwards. Vercel, Render, Railway, Netlify and Livops all ship one.
- Deploy from Claude Code: connect Livops over MCPRun one command, claude mcp add --transport http livops https://livops.uz/api/mcp, then sign in through /mcp. From then on Claude Code can list your projects, deploy, read the build log, explain a failure and roll back, as you and only in the team you picked.
- Deploy from Cursor: connect Livops over MCPAdd Livops to ~/.cursor/mcp.json with just its address, https://livops.uz/api/mcp, and sign in through the browser. Cursor's agent can then deploy, read build logs, explain a failure and roll back, and it asks you before each tool call unless you change that.
- Deploy from Codex CLI: connect Livops over MCPAdd a [mcp_servers.livops] table with url = "https://livops.uz/api/mcp" to ~/.codex/config.toml and run codex mcp login livops. Codex can then deploy, read build logs, explain a failure and roll back, and with default_tools_approval_mode = "writes" it asks before anything that changes.