What gets saved
A Custom Agent saves a way of working.
A Custom Agent is a reusable set of instructions for a particular writing or workflow job. It can hold tone, format, target language, guardrails, shortcuts, and—in custom write mode—ready PDF, TXT, or Markdown knowledge files.
Custom AI agents for Mac are most useful when each one has a narrow job. “Write like me” is difficult to judge; “turn my rough weekly notes into a concise founder update with progress, blockers, and next steps” gives the system and the user a much clearer standard.
The same principle applies to workflows. An Agent can help a recurring task use the right format and constraints, but it does not grant permission to send a message or change existing third-party content. Personalization should make results more consistent, not make actions less visible.
What an Agent should remember
Save the stable rules. Supply the changing facts.
The best division of labor is simple: the Agent keeps recurring preferences; each new request supplies the current situation.
Writing rules become reusable
A writing Agent can keep a familiar voice, preferred structure, terminology, target language, and response boundaries when turning rough input into a draft.
Workflow output gains a consistent shape
For repeat tasks, an Agent can define how a note, plan, checklist, or draft should be organized while the workflow handles the current details and supported destination.
Reference material can stay attached to the job
Custom write Agents can use processed PDF, TXT, and Markdown files as context. A file must be ready before it can ground the output, and the user should still verify source-sensitive claims.
Build with a real job in mind
A good Agent is specific enough to test.
Name the recurring situation
Start with a task you already repeat. Describe the audience, the expected result, and what a good answer must never change or invent.
Add only the rules that earn their place
Set tone, length, language, format, guardrails, and a shortcut when they make the output easier to judge. Attach knowledge only when the job genuinely needs it.
Test it with imperfect input
Try a real rough note, then check whether the Agent preserved facts, uncertainty, and boundaries. Adjust the saved behavior before relying on it for repeated work.
Two useful patterns
The strongest Agents have a clear editorial job.
A good custom agent captures a repeat rule the user would otherwise explain again and again.
Founder update agent
Turn this rough voice note into our weekly investor update using my usual tone.
A concise weekly update is prepared with progress, blockers, next milestones, and the founder's preferred writing style.
The recurring sections and tone live in the Agent. This week’s progress and blockers still come from the founder’s current notes.
Product workflow agent
Create a launch checklist from these notes and keep it aligned with our release process.
A structured launch checklist is prepared using the agent's saved format and knowledge files.
The release format and ready knowledge files provide context; the checklist remains reviewable rather than being treated as an automatic launch action.
Use the simplest tool that fits
Not every task needs a Custom Agent.
| Use case | Custom agent | Released built-in flow | One-off prompt |
|---|---|---|---|
| Personalization | User-defined rules, tone, and format | Product-defined behavior | Restated each time |
| Knowledge | Can use supported knowledge files | Usually general workflow context | Depends on pasted context |
| Best for | Repeat personal workflows | Common Aureon tasks | Exploration or one-time tasks |
| Control | Reusable but reviewable | Fast default path | Manual prompt control |
Design for trust
An Agent should narrow behavior, not expand authority.
Use a custom agent when the user has a repeat preference that should carry across both writing and workflow output.
Best fit
Repeat rules
- Personal writing tone
- Company formats
- Knowledge-backed writing
Use carefully
Sensitive output
- Legal or medical content
- Recipient-facing messages
- Claims that need source review
Not the job
Unsupported claims
- One-click support for every app
- Unsupported app actions
- Sending or editing without user review
Before creating one
Questions about Agents, files, and control.
Start with one recurring job and a result you can evaluate. Complexity is useful only when it makes that result more dependable.
01Do custom agents belong to AI Input Flow or AI Workflow?
They cover both. A custom agent can guide expression polish and cross-language communication, and it can also guide shortcuts and smart workflows.
02Can custom agents use files?
For custom writing workflows, Aureon can use supported knowledge files such as PDF, TXT, and Markdown so the agent follows the user's own context.
03Are custom agents the same as built-in flows?
No. Built-in flows give users a fast default path. Custom agents let users define their own tone, formats, language rules, workflow behavior, and optional knowledge context.