Skip to content
AppDues
Log in
Contact salesSign up

IntelliJ Plugin for Subscription Apps: AI Agent, Paywalls, and Metrics in Android Studio

Subscription work often pulls Android developers out of the IDE: one browser tab for MRR, another for offerings, a third for paywall design. An IntelliJ plugin for subscription apps brings those tasks into the tool window. It pairs a dashboard panel with an AI agent that can read your project, propose source edits, create subscription resources, start paywall generation, and query revenue charts from natural-language prompts. This guide explains what each part does, where human review still matters, and how to phrase requests that produce reviewable results.

Sign-in and permission scope

The plugin authenticates with OAuth 2.0 using the PKCE flow instead of a manually pasted API key. On first open, the tool window shows a welcome screen with a sign-in button. Choosing it launches your browser; the plugin runs a local callback server, completes the token exchange, and stores credentials. After authorization it loads your projects so you can pick one.

The requested scopes cover project configuration, charts and metrics, and customer information. Tokens refresh automatically when they expire, which keeps you signed in across sessions without re-entering anything. Treat the scope list as the boundary of what the agent can touch: it is a project-management and analytics assistant, not a general-purpose account administrator.

How the agent edits your source files

The agent can read, search, and modify project files, so requests such as adding SDK initialization to an Application class or changing the offering ID on a paywall screen are within reach. The important control is that proposed changes arrive as staged edits with an inline diff. You can accept one edit while rejecting another, accept the whole batch, or discard everything without touching your files.

Use staged review as a gate, not a formality. Check that a generated change matches your actual class names, dependency versions, and offering identifiers before applying it. The diff preview exists precisely because an agent's suggestion can be plausible but wrong for your codebase.

End-to-end paywall generation

A single prompt can cover the whole paywall setup chain: create the offering, add packages, attach products, and generate a designed paywall with copy, images, styling, and a template choice. The agent monitors the generation job in the background and, when the design is ready, shows a notification with a link into the paywall builder for review and publishing. Monitoring continues even if OAuth tokens expire, because an API key acts as an independent fallback.

Multiple jobs can run in sequence. Asking for a premium paywall and then a freemium paywall is handled as two tracked jobs without conflict. Generation produces a draft for review; publishing remains a separate action you take in the builder.

Managing offerings, entitlements, and paywalls by conversation

The agent supports create, read, update, and delete operations across subscription-platform resources. Useful prompt patterns include:

  • Creating an offering with packages and prices, for example a Premium offering with a monthly package at $9.99 and an annual package at $79.99. If the resources already exist, the agent detects duplicates and reuses them rather than failing.
  • Creating an entitlement such as pro_access and attaching it to all subscription products. The agent creates the entitlement, lists existing products, and attaches each one.
  • Auditing configuration: asking which offerings have paywalls, or which products are not attached to any package. The agent cross-references products, packages, and offerings to surface orphaned resources.
  • Generating a paywall with app context. A request that describes the app, its audience, and its offering gives the agent material for relevant copy and styling.
  • Duplicating a paywall under a new name, with a builder link returned for customization.

Specific prompts beat vague ones. Naming the offering, the package cadence, the price, and the intended audience gives the agent concrete inputs and makes the result easier to verify.

Revenue questions and integration debugging

The agent can query 21 chart types, including revenue, MRR, ARR, churn rate, trial conversions, active subscriptions, and retention cohorts. It translates a question into Charts API parameters such as date ranges, resolutions, and segment filters, then presents the result readably. Example questions: the MRR trend over six months, weekly trial-to-paid conversion, this month versus last month revenue, or churn among annual subscribers.

Beyond data retrieval, the agent can diagnose integration problems. For a 'Configuration not found' error when presenting a paywall, it checks project configuration and verifies that offerings have packages with products attached to find the root cause. It can also walk through SDK setup for a platform such as Kotlin Multiplatform using Codelabs content, and explain concepts like offering lookup keys versus offering IDs in the context of your actual configuration rather than generic documentation.

Choosing a model

A gear icon in the input bar switches between models: GPT-4.1, GPT-4.1 Mini, Claude Sonnet 4, and Claude Haiku 4.5. The trade-off is capability versus response speed. Use the lighter options for quick lookups and routine edits, and the fuller models for multi-step generation or debugging where reasoning quality matters more than latency.

What the dashboard panel shows

The dashboard panel gives a live project view without a browser. Its collapsible sections load when the tool window opens.

  • Revenue overview: active subscribers, active trials, MRR, and revenue, pulled from the subscription platform API.
  • Milestones: the same achievements shown on the dashboard, including first subscriber, revenue thresholds, and trial milestones.
  • Configured offerings: every offering with its packages and attached products; expand one to see package structure, attached products, and whether a paywall exists. Offerings with paywalls link to the builder.
  • Paywalls: all paywalls with colored status badges — Published (green), Draft (yellow), and Has Unpublished Changes (orange). Action buttons depend on status: Edit, Publish, Unpublish, Discard, Duplicate, and Delete. Edit opens the builder in your browser; lifecycle actions call the API and refresh the panel automatically.

A review checklist before you apply or publish

  1. Confirm the target project is the one you intend; the agent acts on the selected project.
  2. Read each staged diff and check identifiers, imports, and dependency assumptions against your codebase.
  3. For generated paywalls, open the builder link and review copy, images, and template before publishing.
  4. For resource creation, verify that duplicate detection reused the right existing offering or product rather than creating a near-duplicate.
  5. For chart answers, check the date range, resolution, and segment filters match the question you asked.
  6. Use the dashboard panel's status badges to confirm whether a paywall is published, draft, or has unpublished changes.

Frequently asked questions

Does the plugin still need a manually configured API key?

Sign-in uses OAuth 2.0 with PKCE, replacing the older manual API key configuration. An API key still serves as an independent fallback for monitoring paywall generation jobs when OAuth tokens expire.

Can I reject part of an AI-proposed code change?

Yes. Proposed changes appear as staged edits with an inline diff. You can accept individual edits, accept all, or reject the batch without modifying your files.

Can the agent generate more than one paywall at a time?

It can run multiple generation jobs in sequence, such as a premium paywall followed by a freemium paywall. Each job is tracked independently, and monitoring continues even if OAuth tokens expire.

What revenue questions can I ask?

The agent queries 21 chart types covering revenue, MRR, ARR, churn rate, trial conversions, active subscriptions, and retention cohorts. It applies date ranges, resolutions, and segment filters based on your question and presents the results readably.

Which AI models are available?

GPT-4.1, GPT-4.1 Mini, Claude Sonnet 4, and Claude Haiku 4.5. Switch between them with the gear icon in the input bar to balance capability and response speed.