Page
Work · Project
Subject
Apple Search Ads agent
Role
Builder · Solo
Status
OPEN SOURCE
PROJECT.04 · Project OPEN SOURCE Agentic AI Growth 2025 → ongoing

A semi-automated agent for Apple Search Ads, with you in the loop on anything that spends.

It pulls performance, scores keywords, mines search terms, and proposes the next moves. You approve anything that touches budget; it handles the mechanical work in between.

Apple Search Ads agent memory A keyword's PKEI score plotted over six cycles (solid green line, "meditation") against a candidate alternative (dashed slate line, "stress relief"). When the alternative overtakes, the agent swaps to it; when the original recovers and scores higher again, it reverts. Each decision is recorded in an append-only ledger beneath the chart, with its PKEI score and the human's approval. MEMORY · APPLE SEARCH ADS OPEN SOURCE PKEI high low SWAP REVERT meditation stress relief C1 C2 C3 C4 C5 C6 LEDGER · append-only 04-26 SWAP "stress relief" PKEI 7.1 ✓ you 05-10 REVERT "meditation" PKEI 7.4 ✓ you EVERY CALL LOGGED, ACROSS MARKETS HOLD · SWAP · REVERT
What it is

In one paragraph.

A small agent for managing Apple Search Ads, built for StillMind and open-sourced as a starting point. It is semi-automated: autonomous on the mechanical work, pulling data, scoring keywords and drafting recommendations, and human-in-the-loop on anything that spends. Clone it, point it at your account, and shape the workflow and the scoring to your own app.

The two ideas

What makes it different.

USP.01
Append-only campaign ledger

Memory that compounds

Every change is logged with the reasoning behind it, the score that justified it, and your decision, then matched against how the keyword performed afterwards. So a call is never made on a single snapshot. You can read a keyword's whole history across markets and decide to hold it, swap it, or switch back to the one that was quietly working better.

USP.02
Tunable PKEI

Scoring you can tune to your app

Keywords are ranked by one transparent score (PKEI) rather than opaque model judgement, and the formula is yours to tune. Calibrate its thresholds to your app's authority, maturity, and traffic so a young app and an established one each prioritise the keywords that matter at their stage.

What's in the box · 5

The features that matter.

Feature.01

Analyse performance

Pull campaign, keyword and search-term data on demand, then rank every keyword by PKEI to see what is earning its spend.

Feature.02

Tune and negate

Adjust bids, pause keywords, and negate the junk search terms quietly burning budget, within the guardrails you set.

Feature.03

Plan and add keywords

Mine converting search terms, graduate them to exact match, and plan new keywords before they go live.

Feature.04

Create and manage campaigns

Spin up campaigns and ad groups (always paused for your review) and run them across multiple countries and currencies.

Feature.05

You approve the spend

Anything that moves money (a bid, a budget, a new campaign) waits for your sign-off, with the full rationale attached.

Why I built it

The motivation.

Running Apple Search Ads properly is hours of mechanical work every week, and most of the decisions inside it are rule-following with the odd judgement call. That is ideal agent territory: automate the rule-following, surface the judgement, and keep a record of every call so the next decision is better than the last.

Status & links

Where to find it.

Current status
OPEN SOURCE Apache-2.0 · on GitHub