AI Agent

An assistant that knows your projects: ask it to deploy, explain a failed build or find out why an app is slow.

The StackBlaze agent works on your account the way you would: it reads your projects, apps, builds, logs and metrics, and it can deploy and change things for you. It runs with your own permissions, so it only sees what your role can see. You can use it in the dashboard or from the terminal.

In the dashboard

Click Ask Agent in the header to open the chat panel on the right. The agent knows which project, environment, app or add-on you are looking at, so "why did the last deploy fail?" works without naming the app. Type @ to attach another project or app, and paste or drop a screenshot to show it an error.

You can also start from the home page ("What do you want to get done today?"), or press Diagnose on an app's metrics or an add-on's deployments to ask why it is slow or failing.

Things to ask

Goal	Example
Deploy	"Deploy github.com/acme/shop with a Postgres database"
Explain a failure	"Why did the last build of api fail, and how do I fix it?"
Diagnose	"Why is @web slow since this morning?"
Change config	"Set LOG_LEVEL=debug on the worker and redeploy"
Find things	"Which apps in shop use more than 80% of their memory?"
Templates	"Set up n8n with a database"

It asks before risky actions

Reading is instant, but actions that can break or remove something wait for you. The agent shows a confirmation card, and nothing happens until you approve it. That covers deleting an app, project or function, restoring a snapshot, restarting or scaling an app, resetting a database password, and opening a support ticket for you.

Warning

Deleting an app also deletes its volumes and the databases created just for it. Read the confirmation card before you approve.

From the terminal

The CLI has the same agent. Run it without a prompt for an interactive chat, or with a prompt for a single answer, for example in a script:

terminal

stackblaze agent
stackblaze agent "why did my last deploy fail?" -a api
stackblaze agent --json --prompt "list my apps" | jq -r .response

In the interactive chat, /project, /env and /app change the scope, and /help lists the rest. Without a terminal, destructive actions are declined unless you pass -y. See agent in the CLI reference.

In your own AI tools

Prefer Claude Code, Cursor, Codex or VS Code? Connect them to the StackBlaze MCP server and they get the same StackBlaze tools. stackblaze setup agent -y sets up every AI tool it finds on your machine.

Good to know

Topic	Details
Permissions	The agent acts as you. Making changes needs write access to the app; read-only members can ask questions but not change anything.
Databases	The agent can query your database add-ons. In the production environment it only reads them; in other environments it can also write, if you have write access.
Plans	The agent is part of the current plans. If your plan or your platform does not include it, the panel says "AI features are not included in this plan." or "The chat assistant is disabled on this platform."
Usage	AI usage is metered per organization and shown on the Billing page.
History	Dashboard conversations are kept in your browser (up to 50). Use "New conversation" to start fresh and "Conversation history" to go back to one.
Accuracy	The agent can be wrong. Check what it proposes, especially before you approve a change in production.