Documentation

Getting started

Install the app from the Shopify App Store. OAuth runs automatically and the dashboard opens with the Free plan (10 products per period, full feature set). Already installed? Open the app from Shopify Admin → Apps — the embedded app runs at app.baifpo.com.

1. Connect an AI provider — choose one of the eight supported providers and paste your API key (see provider setup guides below).
2. Choose your model — pick your provider first, then a model. Quality tiers, published prices and custom model ids are shown in the app.
3. Pick the products — use the product picker: search, filter, load 10 per page, select recommended priorities, or select exactly the products you want.
4. Review and apply — inspect the grounded improvements, then Apply them to Shopify or copy a field. Every analyzed product uses one plan slot; rewrites scoring 90+ are marked as counted results (up to 3 per product).

Understanding findings

Each finding explains what is wrong and why it matters — for example a generic title, a thin description or missing tags. Findings are grouped by product, and results are labelled:

• AI-assisted — the AI enrichment used your provider's model on your key. Results show the model, the AI calls used and whether the result counted toward your plan.
• AI error / limit reached — the product was screened by the built-in quality rules instead of a new AI call. Retry it later, or switch to a different or stronger model and use “Refresh with current AI model”.
• AI checked — no changes needed — the model reviewed the product and suggested no rewrite.

Applying or copying improvements

You choose what happens to each improvement, and nothing is written until you click:

• Apply to Shopify — writes the improved title, description and/or tags to that product. The app uses the write_products scope only for this action, only for the fields shown, and only after you confirm. Applied products are marked and stay visible in Catalog Intelligence with a before → after score.
• Copy — copies a single improved field to your clipboard so you can paste it wherever you prefer.

Grounded improvements are validated before you see them: numbers and brand names are preserved, invented specifications and claims are rejected, and HTML is sanitized. If a product looks unchanged after applying, refresh the picker — it reloads live data from Shopify.

What BYOK means

Bring Your Own Key. You supply your own AI provider API key. AI requests are sent to the provider you configure, and that provider bills you directly. The app:

• encrypts your key at rest and never returns it to the browser;
• never logs your key;
• adds no AI markup and sells no AI credits;
• never silently switches you to another provider.

Supported providers

BAIFPO is BYOK: connect your own API key. Eight providers are supported: OpenRouter, Google Gemini, OpenAI, DeepSeek, Anthropic, Requesty, NanoGPT and Eden AI. Each provider uses its own official API and key format; free tiers and limits are controlled by each provider and can change.

• OpenRouter — many models, one OpenAI-compatible key. Free models may be available.
• Google Gemini — Gemini 3.x Flash models; free tier may be available.
• OpenAI — GPT models including the GPT-6 family.
• DeepSeek — low-cost reasoning models.
• Anthropic — Claude models including Opus 5.5 and Sonnet 5.5.
• Requesty — many models through one OpenAI-compatible key; free models may be available.
• NanoGPT — free to start: add balance from $1 (or $0.10 with crypto) and pay per token. One free model is available on the web only; many open-weight models are included in the monthly subscription.
• Eden AI — one API for hundreds of AI models; free models may be available.

Data usage by providers. BAIFPO never uses your store data to train or fine-tune AI models. What a provider does with the content you send is governed by your provider account and its terms — data-use and retention policies can differ between providers, and between paid and free tiers, so check your provider's data-use policy before sending sensitive content.

Choosing a model and estimating costs

The app keeps a maintained catalog of models with quality tiers and published prices. After connecting a provider:

• Model list per provider — from budget models to frontier models (GPT-6, Claude Opus 5.5 / Sonnet 5.5, Gemini 3.8 Flash, Grok 4.7 and more). The current list for each provider, with quality tiers and published prices, is shown in the app — the catalog changes over time, so we do not copy the full list here.
• Quality-first recommendation — for content optimization the app recommends a high-quality model and shows alternatives. You can always pick any model you prefer.
• Use any other model from this provider (advanced) — if your provider supports models not in our list (for example many OpenRouter models), open this section in the app and enter the exact model id your provider expects. Catalog prices may not apply to custom models; your provider bills actual usage and we cannot estimate it in advance.
• Cost simulator — the default preset estimates this app's product-optimization workload (one AI run per product you analyze), prefilled with your plan volume. Tick "Show other AI workloads (outside this app)" to also estimate translation, marketing copy, SEO title/meta description or brand campaigns you run with the same provider. Adjust tokens and runs to match your real usage and compare estimated monthly costs of switching models.

Where to find model ids. The custom-model field expects the exact API model name (case-sensitive) your provider uses. Copy it from the provider's official model list:

• OpenRouter — openrouter.ai/models (e.g. openai/gpt-4o-mini)
• Google Gemini — Gemini API models (e.g. gemini-3.8-flash)
• OpenAI — platform.openai.com/docs/models (e.g. gpt-4.1-mini)
• DeepSeek — api-docs.deepseek.com (e.g. deepseek-chat, deepseek-reasoner)
• Anthropic — Claude models overview (e.g. claude-3-5-haiku-latest)
• Requesty — Requesty model library (e.g. openai/gpt-6.1-sol, openai/gpt-6-luna)
• NanoGPT — nano-gpt.com/models (e.g. openai/gpt-6-sol, deepseek/deepseek-v4.1-flash)
• Eden AI — Eden AI models (e.g. openai/gpt-6.1-sol, google/gemini-3.8-flash)

Estimates use published provider prices and may differ from your final bill; your provider bills you directly.

Provider setup guides

API key security

Your API key, your provider account. Your AI provider account remains under your control — BAIFPO only stores the key you provide.

• Rotate API keys regularly.
• If you believe a key may have been exposed, revoke it immediately and replace it with a new key.
• You are responsible for maintaining the security of your provider account and API credentials.

• Keys are encrypted at rest (Fernet) and decrypted server-side only for a request.
• Keys are never returned to the browser, never logged, and never included in telemetry.
• Removing a provider deletes its encrypted credential.

Product analysis

The app reads product titles, descriptions and tags with the read_products scope only. It analyzes:

• clarity, completeness and structure of titles and descriptions;
• presence and usefulness of tags;
• duplicates and near-duplicates;
• factual grounding of suggested improvements.

Improved content is validated deterministically: no invented materials, dimensions, certifications or claims; numbers and brand names are preserved; HTML is sanitized.

Catalog Intelligence

Included on every plan, Free included. It is deterministic (no AI cost) and based on your latest analyses:

• Health score with component bars: content quality (45%), content completeness (20%), duplicate control (15%) and freshness (20%). Each component is 0–100, higher is better.
• Priority bands — High / Medium / Low with one-click selection of priorities, duplicates or declined products straight into the picker.
• Check for changes — compares live Shopify content against your last analyses and lets you select changed or never-analyzed products.
• Recently applied — a log of improvements written to Shopify with before → after scores; applied products stay out of the priority ranking and can be re-run for free.

Usage & limits

Every plan includes the full feature set. Every analyzed product consumes one plan slot per period, regardless of how it scores; re-analyzing an already-analyzed product never consumes another slot. A single run analyzes up to 10 products on every plan. Rewrites scoring 90+/100 after optimization are marked as counted results, up to 3 per product. AI runs have a fair-use cap of 6× your plan allowance per period — beyond it, products are checked deterministically. The allowance resets each period.

Billing

Plans are billed monthly through Shopify App Pricing. Upgrade, downgrade or cancel in Shopify at any time — no reinstalls required. After a plan change the app re-verifies your subscription with Shopify before applying the new entitlement. A voluntary 7-day refund for your first paid charge is available — see our Terms.

Security & privacy

• Product access with read_products, plus write_products used exclusively when you click Apply. No orders, customers or analytics scopes.
• BYOK credentials encrypted at rest; never exposed to the browser or logs.
• Tenant isolation: each store only ever sees its own data; uninstalling deletes the store's data.
• Product content sent to an AI provider is subject to that provider's terms and privacy policy.
• See the Privacy Policy and Terms of Service.

Troubleshooting

• Connection failed / invalid API key — check the key, its permissions and your provider account balance. Then test again.
• Provider timeout — the provider took too long; try again or choose another model.
• Analysis failed — retry; failed analyses never consume quota.
• Quota reached — all plan slots for this period are used; upgrade or wait for the next period. (Counted results are a quality label: 90+ rewrites, up to 3 per product.)
• 3 counted results reached — a product already produced three 90+ results this period; it is skipped for further AI runs until the next period.
• Fair-use limit reached — the period's AI runs (6× your plan allowance) are used; remaining products are checked deterministically until the next period or an upgrade.
• Model changed but old result shown — switching models bypasses the stored result automatically; use "Refresh with current AI model" to force a fresh run.
• Apply failed (permission) — the app needs the write_products scope; reinstall/approve access, then try again.
• Billing not updated — reopen the billing page; the app re-verifies the subscription with Shopify.
• After reinstalling — fresh approval is required before a paid plan applies; that is expected.
• Shopify authentication — reopen the app from the Shopify admin to refresh your session.

FAQ