How to Define Custom Extraction Rules for Your Clients

Audience: Fractional CMOs (Agency tier and above)

Time estimate: 10 minutes

Prerequisites: Agency tier or higher, active client mind, some understanding of what context matters most to your clients

What You'll Accomplish#

Define custom extraction rules that teach Coppermind what to pay attention to when ingesting content about your client. Instead of using Coppermind's default extraction logic, you specify the facts, preferences, decisions, and signals that matter most for this particular client or industry. Your custom rules apply automatically to all future ingestion — no manual markup needed.

How It Works#

Custom Extraction Rules store per-client instructions that Coppermind uses during the ingestion pipeline. When you ingest emails, files, meeting transcripts, or other content, Coppermind extracts structured memories using your rules as guidance. The rules live in the client mind's configuration and are applied by the IngestEngine—the system that breaks down raw content into usable memories.

You can define:

• What facts matter — which details should be stored as standalone memories vs. ignored
• What signals matter — which buying signals, objections, or preferences should be tagged separately
• Industry-specific context — if you work with SaaS companies differently than ecommerce, you can define rules per industry
• Decision patterns — how to recognize and flag the client's decision-making style, risk tolerance, budget constraints

Usage#

Custom Extraction Rules are available as the tune tool when you're switched to an Agency-tier client mind.

Step-by-Step#

1. Switch to the client you want to define rules for
2. Call tune with action="get" to see current rules (if any)
3. Define your rules using natural language descriptions of what matters for this client
4. Call tune with action="set" to store your rules
5. Ingest new content — future ingestion will use your rules automatically

Example#

Instead of Coppermind guessing what to extract, you could:

"Set custom extraction rules for Acme Corp. Flag any mention of budget constraints, competitive concerns, or stakeholder concerns about timing. Extract their decision-making style — they move slow and need internal consensus. Extract vendor preferences explicitly."

Coppermind will store these rules and apply them whenever you ingest emails, meeting notes, or files about Acme.

What You'll See#

Custom Extraction Rules returns:

• Current rules (if you call action="get") — the rules currently stored for this client
• Confirmation (if you call action="set") — rules saved and will be applied to future ingestion
• Clear state (if you call action="clear") — previous rules removed, extraction reverts to default behavior

The rules themselves are plain-language instructions stored in the client mind's configuration. You can view and edit them whenever you want.

When to Use Custom Extraction Rules#

Custom rules are most useful when:

• You've worked with the client long enough to know what they care about
• Your industry has specific signals (SaaS buyers care about implementation timelines; ecommerce founders care about logistics)
• You're about to do a large ingestion (email backfill, file upload) and want Coppermind to focus on what matters
• You want to avoid cluttering the memory store with irrelevant facts

You do NOT need custom rules if:

• You just started working with the client (let default extraction run first, see what Coppermind captures, add rules later)
• Your client's signals are already well-captured by Coppermind's defaults
• You're ingesting ad-hoc content occasionally (the overhead isn't worth it)

Troubleshooting#

"I got an access denied message"

You're on a Pro or Solo tier. Custom Extraction is an Agency-only feature. You'll see a preview of what custom extraction includes plus a link to upgrade to Agency.

"My rules don't seem to be applying"

Custom Extraction Rules are applied during ingestion, not retroactively. Rules you set today apply to new content you ingest tomorrow. If you want to re-extract old content with new rules, you can use the documents tool to re-extract from raw documents already stored.

"How do I know if my rules are working?"

After setting rules and ingesting new content, use search_memory to look for the facts or signals you defined. If you set a rule to "flag budget constraints as buying_signals," search for "budget" and check that new memories are tagged as buying_signal type. If they're not, adjust your rule language and try again.

Next Steps#

After defining custom extraction rules:

• Test with a small batch — ingest one email thread and verify the rules picked up the right signals
• Iterate if needed — if Coppermind is still missing things, refine the rule language and re-ingest
• Document why — add a note in the client's context about what this rule is for (e.g., "they move slow on decisions—always flag timeline concerns")
• Maintain as you learn — as you work with the client longer, refine your rules to match new learnings
• Share with the next CMO — when you hand off, the rules transfer with the client mind. The next CMO inherits your extraction logic.

Tip: Start with 3–5 focused rules rather than trying to capture everything. The best extraction rules are specific enough to catch signal noise but broad enough not to miss nuance. Example: "Any mention of budget or spend limitations" is better than "Mention of anything related to money."