Every Lyra tool, one chat.
AI Advisor is the chat surface of the Lyra Agent. The Agent reads your real account data, answers strategy questions, and when you ask for an action it dispatches the matching Lyra tool — bid optimisation, campaign creation, health analysis, change-history alerts. Read-only by default; nothing writes to your account without you saying yes.
Your account, in scope. 52 tools, ready to call.
AI Advisor reads from three places in your Lyra workspace and can dispatch any of 52 registered functions. It never invents context; everything in its responses is grounded in something you've configured or written.
What it looks like in chat.
Four real-shape scenarios. The advisor reads your account, reasons about the question, dispatches the matching Lyra tool, and waits for your approval before anything writes. Click a chip to switch.
What the advisor reads, every turn.
Three DB sources ground every advisor response. No third-party data, no synthesised competitive intel, no scraped benchmarks — what you've stored is what the advisor reads.
The same data your dashboard shows.
customer_cache holds the live account state — connected customers, campaigns, conversion data, last-sync timestamps. The advisor pulls from the same table the customer-portal dashboard reads, so the numbers it cites are the numbers you see.
Source · customer_cache · refreshed by data-sync engineEvery decision you've logged is in scope.
account_diary_entries holds the running record of decisions and rationale — what you paused, why, when. The advisor reads these for context, so questions like "remember when we restructured Brand · Search?" land somewhere real.
Source · account_diary_entries · per customerHard constraints the advisor obeys.
customer_onboarding_contexts holds what you told us at setup — target ROAS, brand-protection rules, locations to never exclude. The advisor treats these as constraints; if a recommendation would violate one, it's flagged before the chat suggests it.
Source · customer_onboarding_contexts · 14 active rulesA chat, on a short leash.
Every response is text plus an optional dispatch suggestion. Tool execution requires your explicit yes. Read-only at the model level — the Agent can't bypass the approval gate.
Three DB sources; nothing else. No third-party benchmarks, no scraped competitive intel, no synthesised numbers. What's in your workspace is what the advisor knows.
Bid optimisation, campaign creation, health analysis, change-history activation, search-terms management, and 47 more. The function table is the agent's tool surface — if a capability isn't there, the advisor can't run it.
Every chat is saved in chat_conversations. Resume tomorrow, share with a teammate, audit later. Nothing is shared across customers or accounts.
The honest answers about the chat.
Can the advisor change my Google Ads account on its own?
No. Every action that mutates Google Ads — bid changes, campaign creation, applied recommendations — runs through an explicit approval step. The model cannot bypass it; the gate is enforced in the dispatch layer, not the prompt. Whatever the advisor proposes, you approve before it ships.
What's actually powering the chat?
The Lyra Agent — our name for the AI layer across the product. Today it runs on Claude Sonnet from Anthropic, via the function-calling pattern. The Agent receives your message, decides whether to read more from your DB, optionally proposes a function call, and returns text plus an optional dispatch suggestion. The orchestration code is ours; we may change the underlying model as the field evolves — the contract with you is the Agent's behaviour, not the model behind it.
What can the advisor not do?
Anything that doesn't have a registered function. The 52-function table is the boundary — if a Lyra capability isn't there, the advisor can't run it. New functions get added as we ship new tools.
Does the advisor remember our last conversation?
Yes, for 30 days — every conversation is saved in chat_conversations. The advisor reads recent history alongside your diary to answer follow-ups in context. After 30 days the chat is archived; the diary entries it produced stay in place.
Why use the advisor instead of running the tools directly?
Speed and synthesis. The advisor reads your account before answering, so you skip the pull-metrics → form-question → find-the-right-tool cycle. For an experienced operator, going direct to a tool can still be faster — both paths exist.
Run the audit.
Keep the findings.
Connect your account and let Lyra run its 18 tools for 14 days. If the projected waste recovery isn't worth at least 10× the $49, don't pay. You keep every insight either way.
Read-only connect · write access opt-in per tool · SOC 2 in progress · GDPR + CCPA compliant