Packages
Save a task with multiple steps as a package.
Run it when needed, and repeat the same work if you choose.
Understanding packages
A package is a sequence of tasks for an AI to follow. It groups work such as “summarize meeting notes → identify action items → save a file.” Each time it runs, the AI carries out the saved sequence.
| Role | What it does | Example |
|---|---|---|
| Authoring AI | Talks with you to build the task sequence. | “How should we organize the meeting notes?” |
| Execution AI | Actually carries out the saved sequence. | Summarizes notes and writes a file. |
“Create with AI” builds the task sequence. A saved package starts doing the actual work when you press “Run.” You can choose different AIs for authoring and execution.
Example: a package for organizing meeting notes
For your first exercise, try a small task that needs no external service connections. Use these notes to create a summary file and action-item list.
Describe the work in a new package
Create a new package in the Packages window and enter the task. You can use the example below as a starting point.
Notes: The sign-in screen is complete. Alex will review the error messages by Friday. Jordan will draft the user guide by next Monday.
1. Summarize the key points in no more than three sentences.
2. Create an action-item list with owners and deadlines.
3. Save it as meeting-summary.md in the working folder.
✓The task and desired output file are clearly specified.
Choose an authoring AI and execution AI
The authoring AI builds the sequence with you. The execution AI carries it out later. For your first exercise, you can use the same AI you have already set up for both roles.
Press “Create with AI” to start a conversation with the actual AI. Ask for changes, such as “Combine summarizing and identifying action items into one step.”
✓Actual steps appear in the task sequence beside the conversation.
Review the sequence, give it a name, and save
Check what it reads, what it creates, and where it saves the result. Name it “Organize meeting notes” and save it.
The draft is the work still being authored. The saved task content is what will run. A draft may remain if you leave while building the sequence with the AI.
✓The package list shows its name and saved steps.
Run it once with repeat turned off
For this first exercise, turn repeat off and press “Run.” In “Progress,” check the current step, elapsed time, and the AI’s execution history.
Pressing Run does not mean the task has completed. Follow its status until the steps finish.
✓The running step changes to a completed state.
Check the output file and run history
Open meeting-summary.md in the results area and check the summary, owners, and deadlines. “History” shows what happened at each step.
Multiple runs can share a working folder. When reviewing a past run, also check whether the file was created by that run.
✓The file content matches the original notes.
Execution AI · 2 saved steps · Repeat off
This example illustrates the workflow. Actual steps and output depend on how you configure the package.
Turning repeat off and stopping are different
| Action | Current work | What happens afterward |
|---|---|---|
| Turn repeat on | Continues the current sequence. | Starts the next cycle after completion. If a schedule exists, its conditions apply. |
| Turn repeat off | Finishes the current cycle. | Ends repetition after this cycle. Waiting for a retry is also part of the current cycle. |
| Stop | Requests cancellation of the current work. | Wait for the stop outcome to be confirmed. Changes already made to external systems are not automatically undone. |
When work stops or goes wrong
| What you see | Checks and next action |
|---|---|
| Waiting to retry | Check the failed step’s error, sign-in, and connection. Fix the cause, then retry the failed step from the previous run. |
| Waiting for a decision | Check whether external changes have already been applied, then choose to run again, mark complete, or skip the cycle as appropriate. |
| Edits do not change the running task | The current cycle uses the sequence it started with. A normal save applies from the next cycle. To change it now, use “Start a new run with these changes.” |
| No output file | Check whether the step completed and whether the run history contains an error. Specify the save location and file name clearly in your instructions. |
Common points of confusion
When should I use a package instead of a regular AI session?
A regular session is fine for a one-time question or exploration. Save a package when you want to repeat the same sequence or handle a task in multiple steps.
Are Hermes scheduled tasks the same as packages?
They appear in the same window, but they are separate scheduled tasks managed by Hermes. An orke package’s execution AI and repeat settings do not apply to them as-is.