Skip to main content

Safety model

AdCue is designed so AI clients can help without accidental live spend. These rules apply across the V2 tool surface.

Core rules

Read vs write

The tool catalog marks each tool with Read-only and Destructive columns from the registry export.

Safe first workflow

  1. get_connections_status or list_connected_accounts
  2. list_campaigns / get_campaign_structure on a known customer_id
  3. Analysis tools (analyze_wasted_spend, analyze_search_terms)
  4. Only then create or mutate — still PAUSED / confirmed
Example safe create argument shape:
Confirm in Google Ads that new structure is PAUSED before enabling. Example destructive call (fails without confirm):

What AdCue will not do

  • Spend budget without a write you (or your client with your approval) initiated
  • Access Gmail, Drive, or non-Ads Google data via the Ads connection
  • Document or expose Meta / multi-platform tools as current product surface
  • Bypass auth or invent fake paid entitlements when billing is unconfigured

Auth boundaries

Sign-in OAuth and Ads OAuth are split on purpose. Connecting ads does not enlarge login scopes. More: Security.

Production hygiene

  • Prefer test or low-spend customers when learning create tools
  • Keep a human in the loop for budget increases and ENABLE transitions
  • Rotate API keys if a client is compromised (Keys)
  • Disconnect Google Ads from Connections to revoke Ads tokens from the product side