AI Agents¶

Adwize uses AI in two ways: an assistant you can talk to, and agents that work in the background around the clock. Together, they keep your data trustworthy with minimal effort on your part.

AI Assistant¶

The AI Assistant is a chat interface available on every page in Adwize. It understands your setup — event types, data sources, existing rules, and integration status — and can take actions on your behalf. You can use it to create and manage monitoring rules, investigate alerts, check the health of your integrations, and get recommendations for improving your monitoring coverage.

Click the floating chat button in the bottom-right corner of any page. The assistant already knows your setup, so you can jump straight to what you need.

Example Prompts¶

Monitoring and rules

• "What should I be monitoring?"
• "Alert me if there are more than 50 errors in 15 minutes"
• "Notify me if page view traffic drops by 40%"
• "Disable the purchase threshold rule"

Alerts

• "Show me recent critical alerts"
• "What alerts fired today?"

Integrations

• "How are my integrations looking?"
• "What did the last audit find?"

Data inspection

• "Show me a sample of purchase events"
• "How many events came in today?"

Tips¶

Background Agents¶

While the assistant responds to your questions, background agents work continuously behind the scenes. These agents monitor your data quality, audit your connected integrations, detect anomalies in your event streams, and surface issues before they affect your business decisions.

Background agents run automatically — you don't need to trigger them. When they find something, the results are available through the assistant, the dashboard, or direct alerts depending on severity. This means problems are caught early, often before anyone on your team notices them.

Optimization roster¶

Insights also runs a versioned agency bench. Each desk is a catalog agent you pin on the tenant (kind@version plus enabled_plays). Catalog plays are priors — examples, impact bounds, and the execute allowlist — not the only legal thought. A specialist that sees something real in the digest can write an inbox observation (play: observation:<slug>). Observations are recommend-only: no execute, no claimed lift. enabled_plays still gates which catalog plays can run.

The live demos are:

• Analytics Strategist — GTM lead on 8724. Maps a broken event to the tag and trigger, then proposes a governed workspace diff.
• Media Strategist — audiences, account config, structure, spend, and new-platform trials. Live faf5 only enables the Customer Match switch.
• Lifecycle Strategist — email, SMS, flows, and list hygiene.
• SEO Strategist and CRO Specialist — Search Console, landing funnel, and a shared page inspect. Pinned on a855e4c0. Recommend-only until a CMS worker exists.
• Forecast Analyst — calibrates impact ranges. Policy-only. Does not invent plays.
• Growth Lead — merges specialist memos into the inbox. Does not execute.

Creative Strategist is a catalog stub and is not pinned. Who the client is comes from a shared client_context, not from a hard-coded industry in the prompt.

Trust and Transparency¶

• Review before confirming — The assistant generates configurations you review before anything is created. You can edit parameters before saving.
• Confirmation for destructive actions — Deleting or disabling rules always requires your explicit approval.
• No black boxes — Audit results include what was checked, what was found, and exactly what to do about it.