Workspace instructions

Workspace instructions tell AI how to work with you. Think of them as the onboarding doc you’d give a new assistant on their first day: who you are, what goes where, and how you like things written. Banter’s assistant reads them before acting, and so does any AI you connect over MCP.

There are two levels:

• Workspace instructions — apply everywhere. Edit them in Settings → Instructions.
• Folder instructions — rules for one folder. Right-click a folder and edit its instructions.

What to write

Keep it short and concrete:

• Who you are and what you use Banter for
• Where important things live (“customer work goes under /Work, one folder per company”)
• How you like notes written (“short bullets, keep names, dates, owners, and next steps”)
• Standing rules (“when summarizing customer calls, always pull out pain points and next steps”)

A starter template

# About me
- Name: …
- Role / company: …
- I use Banter for: customer calls, project notes, follow-ups

# How I like notes
- Short bullets over long paragraphs
- Always keep dates, names, owners, and next steps
- Don't invent details that aren't in the source

# Where things go
- Customer work: /Work/<company>, one folder per company
- Meeting notes are named after the meeting ("Discovery call", "Security review")

Folder instructions: rules with a smaller blast radius

A folder’s instructions only apply inside that folder — which is where the precision lives. The /Work/Acme folder can say “this account is in a security review; always check the latest call before answering,” and only Acme conversations carry that rule.

When an AI enters a folder, it gets the folder’s instructions automatically — yours plus any parent folders’.

What good instructions look like

• “File sales notes under /Work by company name.” ✅ — concrete, checkable
• “When drafting follow-ups, write in my voice: short, direct, no exclamation marks.” ✅
• “Make everything good.” ❌ — nothing can follow this
• Ten paragraphs of biography ❌ — burying the rules

Start with five lines. Add a rule each time an AI does something you’d have done differently — that’s the maintenance loop, and it compounds.