Get a notification when something needs your attention, or let an agent investigate and fix it on its own.
Run automations in one project, across all of them, or with no project at all, on a cloud model or a free local one through Ollama.
Let the routine checks run themselves.
You choose what the automation reads and ask it typed questions. A classification model answers with numbers, not text: a probability for yes or no, a probability for each option, or a score on your scale. Your rules act on those answers, and an agent starts only when a rule tells it to start.
Read a status export or a config every couple of hours, and wake an agent when disk, certificates or capacity need attention.
Every weekday evening, read everything since the last tag, decide whether any of it matters to users, and get release notes drafted.
When a commit adds a package, check its licence, how well it's maintained, and whether you already have one that does the job.
Catch a key, token or password in the latest commit, and plan how to remove it from history and rotate it.
Notice changed functions no test reaches, or a migration that drops a column the code still uses.
If a file or a read-only command can show it, an automation can keep an eye on it.
Save an automation in the project to share it with your team through the repository. Save it as global to run it in every project, in the projects you pick, or with no project.
| Reads only | It can only run commands that read, never ones that change things |
| Your prompt | An agent it starts does what your prompt says, in the project itself |
| No project | Runs in its own empty folder and only sees the folders you allow |
| Tells you | Results show up in the status bar, with a desktop notification |
| Never missed | If Koda was closed when a check was due, it runs once when you open it |