What you'll accomplish
Custom Extraction Prompts let you teach Armbrain what matters for a specific client or industry. Instead of relying on default extraction rules, you define custom patterns: "For this SaaS client, always flag competitive positioning mentions" or "For manufacturing, extract supply chain observations separately."
When you set custom extraction rules, Armbrain applies them to all future ingested content—emails, files, meeting transcripts—so your memories get richer and more tailored to what you actually care about.
Steps
- Open Armbrain and switch to the client you want to customize extraction for.
- Find the Custom Extraction tool in your Armbrain tools list (Agency-tier only).
- Think about what you want to capture differently. Consider:
- What does this client care about that other clients don't?
- What patterns should I flag that might get missed by default rules?
- What client-specific industry jargon or concepts matter?
Examples:
- For a SaaS client: "Flag all competitive mentions—we track Salesforce, HubSpot, Pipedrive, and Zendesk specifically. Store competitor pricing separately."
- For a healthcare provider: "Extract patient acquisition costs, patient retention rates, and insurance reimbursement trends as distinct memory types."
- For a manufacturing client: "Separate supply chain observations from market observations. Flag raw material cost mentions."
- Set your custom extraction rules using the tool. Describe:
- What types of information should be extracted
- How they should be tagged or categorized
- What makes them distinct from generic memory extraction
- Any specific keywords or patterns to look for
- Apply the rules. Once set, they apply to:
- Future emails ingested from Gmail
- Uploaded files and documents
- Meeting transcripts and notes
- Any future content added to this client's client mind
- Test the extraction. Upload a sample email or file, then search for the new categories to confirm the rules are working as expected.
Updating or clearing rules
If you need to change your extraction rules:
- Use the Custom Extraction tool again to update rules for the current client
- Clear all custom rules and revert to defaults if the approach isn't working
Rules are stored per client, so customizing for one client doesn't affect others.
Examples of useful custom extractions
Sales-focused client:
- Extract all pipeline mentions with stage and deal value
- Flag lost deal reasons separately
- Capture sales cycle length and close rates
Marketing agency:
- Extract campaign performance metrics distinctly
- Flag channel-specific insights (social, email, organic)
- Separate brand positioning observations from tactical execution notes
Nonprofit:
- Extract donor engagement and retention signals
- Flag program outcome data
- Capture fundraising event observations
Troubleshooting
"Custom Extraction isn't available" — This is an Agency-tier feature. Check your subscription level. Pro and Solo tiers don't have access.
"My custom rules don't seem to be working" — Custom rules apply to new content going forward, not retroactively to old memories. If you want the rules applied to existing emails or files:
- Set the custom extraction rules
- Re-upload or re-ingest the old content
- Or manually add memories tagged with your custom categories
"The extraction is still too generic" — Be more specific in your rules. Instead of "competitive mentions," try "specific competitor pricing, feature comparisons, win/loss analysis." Instead of "market trends," try "changes in buyer behavior, new market segments emerging, regulatory shifts."
"I set rules for the wrong client" — Custom extraction is per-client. If you set rules on the wrong client:
- Switch to that client
- Clear the rules
- Switch to the correct client
- Set the rules there instead
Next steps
- Start with one or two key extraction rules rather than trying to customize everything at once
- Monitor your ingested memories for the first week to confirm the custom rules are capturing what you intended
- Refine rules based on what you're actually seeing get extracted
- Share your custom extraction approach with your team so everyone on this account uses the same framework