marketerswiki
Home
Resources
marketerswiki

The open playbook for performance marketers who build with AI.

Resources

  • SimpleCRM
  • Ads OS
  • Ledgeros
  • Tag Manager Engine
  • Skills
  • Playbooks
  • Vibecoding

Company

  • About
  • Privacy Policy
  • Terms of Service

© 2026 marketers wiki. All rights reserved.

Built withClaude Code
Free & open source · no ad platform API keys · runs on your machine

Your ad accounts deserve a second opinion.

Ads OS is a local AI audit tool that analyses Google Ads and Meta Ads CSVs, flags what is bleeding, and tells you exactly what to do about it. It learns from your own account history over time and improves its own prompt automatically. Powered by Claude Code CLI — if you have Claude Pro or Max, you already have everything you need.

Get Ads OS on GitHub→See how it works
localhost:5000
Ads OS client overview dashboard

The shift

Most agencies are still auditing ad accounts the slow way.

Before

Spending hours in spreadsheets trying to figure out which campaigns are actually wasting money.

→

After

Upload the CSV. Get a client-ready report in 90 seconds.

Before

Noticing a ROAS drop three weeks after it happened.

→

After

Anomaly detection flags deviations the moment you upload new data.

Before

Writing the same audit commentary every month from scratch.

→

After

Claude writes the narrative, the recommendations, and the budget rationale for you.

Before

Generic AI recommendations that ignore your account history and specific goals.

→

After

The learning system calibrates benchmarks to your real data. Context briefings set business type, attribution model, and per-campaign targets before Claude sees a single number.

How it works

From CSV to client-ready report in four steps.

01

Export your CSV

Download a campaign-level report from Google Ads or Meta Ads. No API keys. No OAuth. Just a CSV.

02

Drop it into Ads OS

Platform is auto-detected. Columns are fuzzy-matched and cleaned. Shorthand numbers, currency symbols, messy date formats — all handled.

03

Claude audits it

Ads OS calls your local Claude Code CLI — the same one you already use as a developer. No data leaves your machine. No separate AI API key. Claude analyses campaigns, flags issues, writes recommendations, and builds the narrative.

04

Send the report

Download as PDF or share the HTML directly. Every report is client-ready out of the box — no editing required.

Features

Everything your audit workflow needs. Nothing it doesn't.

AI Audit Reports

Drop any Google Ads or Meta Ads CSV. Ads OS auto-detects the platform, cleans the data, runs Claude against it, and produces a structured HTML and PDF report you can send to a client without editing.

Ads OS audit section showing campaign data and AI copilot

Executive Dashboard

Blended ROAS, MER, total spend, and CPA in one view. P&L waterfall, revenue vs spend trends, and channel split across all your historical uploads — no manual data wrangling.

Ads OS executive dashboard with KPI cards and trend charts

Anomaly Spotter

Every upload is automatically compared against a rolling median baseline. ROAS drops, CPL spikes, CTR fatigue, and spend pacing issues are flagged by severity — before your client notices.

Ads OS anomaly spotter showing critical and warning flags

Revenue Forecasting

Weighted moving average model with seasonality adjustment projects your spend, ROAS, and conversions forward by 7, 30, or 60 days. Claude writes the narrative. You copy and paste it.

Ads OS revenue forecast with trend chart and Claude narrative

Budget Agent

Rule-based budget reallocation with Claude reasoning on top. Set ROAS floors and CPL ceilings per client. The agent tells you exactly which campaigns to pause, scale, or shift budget from — with confidence scores.

Ads OS budget agent with ROAS rules and reallocation table

Also included

Eight more features built in from day one.

Learning System

After every upload, Ads OS recomputes your account benchmarks from real historical data — median ROAS, CPC, CAC, CTR per platform. After 2 uploads the system stops using generic industry numbers and calibrates entirely to your account. Outcome tracking records whether acted-on recommendations actually moved the needle.

Autoresearch Engine

An autonomous prompt optimizer that runs in the developer environment. It takes the current system prompt, runs it against fixed test payloads, scores outputs on 5 binary evals, applies one mutation at a time, and keeps changes that improve the score. When a better prompt is found it gets reviewed and shipped to all users on next pull.

Context System

Every audit is informed by a structured client briefing: business type, account maturity, report recipient, attribution model, campaign type tags, per-campaign ROAS and CPL targets, competitive context, and period notes. Brand-tagged campaigns are excluded from wasted spend checks. Learning phase campaigns are excluded from underperformance flags.

Structured Audit

50+ checkpoints across Tracking, Campaign Architecture, Creative, Cost Diagnostics, and Account Health. Every check scores pass, warning, or fail.

Competitor Intelligence

Scrapes the Meta Ad Library for any brand. Claude analyses each ad: angle, offer, psychology triggers, funnel stage. Pattern summary across all active creatives.

Pixel Health Monitor

11 checks on your Meta Pixel and CAPI setup — event deduplication, event match quality, CAPI freshness. Scored 0–100.

KPI Alert System

Seven alert types fire on every upload — ROAS below floor, zero conversions, CPA spike, CTR drop, CVR drop, spend overpace, spend underpace. Brand-tagged campaigns are suppressed from ROAS and CPA alerts.

Multi-Client Overview

One table. Every client. Last audit date, blended ROAS, total spend, anomaly count, and status. Run all clients in one click. Multi-client summary report with blended metrics and links to each full audit.

Action Plan

Every recommendation from every source — audit, budget agent, anomaly spotter — consolidated into one prioritised to-do list. Mark done with notes. Metric snapshots are recorded when you mark an action done, feeding back into outcome tracking.

Performance Narrator

Three tones: Executive for the client, Detailed for the team, Urgent for the media buyer. Always four sections: This Week in Numbers, What Worked, What Needs Attention, Recommended Actions for Next Week.

Onboarding Wizard

6-step per-client setup: goals, platforms, ROAS targets, budget rules. Generates a 1-page client brief via Claude when you are done.

Multi-Year Historical Data

Every upload is stored with period_start and period_end dates. Upload 36 months of history in one session and the anomaly detector, forecaster, and learning system all use the correct period-ordered baseline — not upload timestamp order.

Requirements

All you need is three things.

1Python 3.9+
2Claude Pro or Max subscription (for Claude Code CLI)
3A Google Ads or Meta Ads CSV export

No API keys. No cloud subscription. No external data sharing.
Everything runs locally on your machine.

Stop doing audits manually.

Ads OS is free and open source. It runs entirely on your machine using your Claude Pro or Max subscription. No extra tools, no API keys, no data leaving your machine.

Get Ads OS on GitHub→

Built with Claude Code · Open source · MIT License