qbash is where you program AI automations. You write them in Python, and qbash hosts and runs them. This 80-second video walks through one Task that reviews a Dependabot PR.
The Dependabot example
Most of the Task is ordinary code. It fetches the changelog and finds where your code uses the package. Deciding whether the upgrade breaks anything used to need a person. In a Task, it's a function call. The Task asks a model one question, gets a structured answer back, and branches on it to comment on the PR or flag a risky upgrade to the team in Slack.
Move the prompt out of the code and it becomes a saved, versioned object. Edit it, and every Task that calls it picks up the change. Saving a Task deploys it. Each change is a revision you can diff and restore. A new PR can trigger the Task on its own, or your coding assistant can call the same Task as a tool.
You can build Tasks like this one for CI failures, bug reports, support email triage, and invoice processing. qbash handles the credentials and integrations.
Tasks covers the runtime, Task triggers and webhooks covers starting a Task from an event, and saved prompts are documented in Deploying, Versions & Logs.
Video transcript
qbash is where you program AI automations. Write them in Python and qbash hosts and runs them. Take a Dependabot PR. Fetching the change log and finding where your code uses the package is ordinary code. Deciding whether the upgrade breaks anything used to need a person.
In a qbash task, it's a function call. You ask the model one question and get a structured answer back. Your code can branch on that answer. Comment on the PR or flag anything risky to the team in Slack. If you move the prompt out of the code, you can save it as a versioned object, edit it, and every task that calls it picks up that change.
Saving deploys it. Every change is a revision that you can diff and restore. A new PR can trigger this on its own. Or your coding assistant can call that same task as a tool. You can build the same shape for countless scenarios, CI failures, bug reports, support email triaging, and invoice processing.
All of the credentials and integrations are handled by the platform. So, write the process in Python and call a model where a step needs judgment. qbash runs the rest. Start free at qbash.com.