Sigil works with any tool your AI agent connects to.

1,400+ properties. Your agent picks the wrong one.

Pipeline stage definitions, field aliases ("ACV" = hs_acv, not amount), exclusion rules for test deals, and metric formulas decomposed into the right HubSpot properties.

Your agent reads transcripts but can't tell a demo from a check-in.

Call classifications (discovery vs. check-in vs. no-show), cross-source mappings to CRM deals, competitor tracker interpretation, and sentiment analysis rules.

Valid SQL, wrong answer. 3 revenue columns and no one documented which is right.

Per-table field context, column disambiguation (which "revenue" column is canonical), join logic with type-casting, fiscal calendar definitions, and dialect-specific cost awareness. Each engine gets its own onboarding flow with the correct SQL dialect, quoting rules, and namespace shape.

"Active" means different things to billing, product, and the board.

Subscription lifecycle definitions (what counts as churned vs. paused vs. past-due), MRR calculation rules, trial-to-paid conversion logic, and revenue recognition context.

"Signup" has 3 definitions in your codebase. The agent picks the one that inflates by 40%.

Event taxonomy (which events represent real user actions vs. system events), funnel step definitions, attribution rules, and cohort membership criteria.

Apollo has no tier field. Your agent queries it like it does and gets garbage.

ICP criteria resolution (tier definitions live in CRM, not Apollo), seniority mapping, cross-source routing rules, and enrichment field disambiguation.

50 custom fields on Opportunity. Your agent uses the deprecated one.

Object relationships, record type disambiguation, custom field context, opportunity stage definitions, and cross-object reporting rules.