Frequently Asked Questions

What AI models are available?

PocketFounder offers three AI models: DeepSeek R1 (the default), Qwen3-Max, and Gemini 2.5 Pro. All three cost 1 credit per generation and produce 3 ideas per run — the difference lies in their reasoning approach, generation pipeline, and the market intelligence they draw on.

How does the generation pipeline work for each model?

DeepSeek R1 and Qwen3-Max use a two-stage pipeline: Stage 1 fetches a real-time trend brief and runs a dedicated Qwen web-search for market grounding, then Stage 2 generates three ideas in parallel (each slot assigned a different business-model type such as SaaS, marketplace, or API service) and audits them against the grounding. Gemini 2.5 Pro uses a single-stage pipeline that leverages its built-in Google Search tool to ground and generate in one pass. Gemini is self-audited; DeepSeek and Qwen ideas are audited by Qwen.

How do the models differ in generation time?

Gemini 2.5 Pro is typically the fastest (1–3 minutes), using Google Search to pull live market data as it writes (6-minute timeout ceiling). Qwen3-Max takes 2–6 minutes — it runs an internal "thinking" phase alongside real-time web search (6-minute ceiling). DeepSeek R1 is the most thorough but slowest (3–8 minutes), spending significant time on chain-of-thought reasoning before writing (8-minute ceiling). All models stream live progress — reasoning character count, content character count, and source fetching — so you can watch generation happen in real time.

Does my model choice affect the quality of ideas?

Yes, in meaningful ways — though all three produce high-quality, actionable plans. DeepSeek R1 produces the most analytically rigorous ideas through an extended chain-of-thought reasoning phase. Qwen3-Max excels at ideas targeting Chinese-speaking markets (mainland China, Taiwan, Hong Kong) and pairs internal thinking mode with live web grounding. Gemini 2.5 Pro generates ideas informed by the freshest data available — its Google Search integration can reference companies and trends published hours ago.

Which model should I choose?

For most users, start with the default DeepSeek R1 for the deepest analytical rigor. Switch to Gemini 2.5 Pro when you need faster turnaround or the most up-to-the-minute market data. Choose Qwen3-Max if you are targeting a Chinese-speaking market or want ideas grounded in Chinese-language community discussions (36氪, 掘金, V2EX, 百度指数, PTT, etc.).

Why does generation sometimes take longer than expected?

Reasoning models (DeepSeek R1 and Qwen3-Max) spend time "thinking" before writing — this internal analysis phase typically accounts for 60–80% of total generation time. You can watch the reasoning character count grow on the generating screen. Real-time web search — Qwen grounding for DeepSeek/Qwen and Google Search for Gemini — adds time proportional to how many sources are retrieved. The pipeline also extends its own deadline a few times when the model is actively producing output, so a long-running generation usually means the model is still making progress rather than stuck.

How many ideas do I get per generation?

Each run produces exactly 3 unique, deeply researched business ideas. Every idea includes a one-line summary, problem analysis, target customer profile, pricing model, tool stack (≥5 tools), 8–10 step daily workflow, 8–10 step setup guide, proof-of-concept test, ≥5 risks with mitigations, confidence rating with reason, personal moat tied to your skills, first-10-customers strategy, and a month-by-month launch timeline.

How do I get more ideas on the same profile?

After your first batch of 3 ideas, you can click "Load More Ideas" once to generate another 3 using the same profile (max 6 ideas total per profile before you start a new one). Each Load More click costs 1 credit. The pipeline records the names of previously generated ideas and explicitly instructs the model not to repeat them.

What languages can I generate ideas in?

English, Simplified Chinese (简体中文), and Traditional Chinese (繁體中文). Switch the language selector in the top navigation at any time — your preference is saved locally and applied to all future generations. Chinese output also swaps the recommended tool stack (Alipay / WeChat Pay, Aliyun / Tencent Cloud) and trend sources (36氪, 掘金, V2EX, etc.) to market-appropriate alternatives.

What does the Confidence rating on each idea mean?

The confidence rating (High / Medium / Low) reflects how well an idea's market claims are supported by the real-time grounding data retrieved during your generation. After generation, an independent auditor model reviews each idea against the trend brief. The auditor downgrades confidence by one level when key claims (whyNow, provenReference) are not traceable to the grounding context, and upgrades by one level only when the grounding contains at least two concrete data points (named products, metrics, or dates) supporting both claims. Never changes by more than one level, and never downgrades Low.

What is the grounding context used during generation?

Before generating ideas, the pipeline fetches a live trend brief for your target region (cached 24 hours) which fans out to 10–29 sources — HackerNews, Reddit, Dev.to, Product Hunt, GitHub Trending, Hugging Face, plus region-specific sources like 36氪, 掘金, V2EX, 百度指数, iThome and PTT for Chinese markets. For DeepSeek R1 and Qwen3-Max it additionally runs a Qwen real-time web search tailored to your skills, interests, and pain points, cached for 10 hours per unique profile. These sources anchor every market claim your ideas make.

What is the "Master Prompt" on each idea card?

Every idea has a "Master Prompt" button that generates a ready-to-paste deep-dive brief you can drop into Claude, ChatGPT, or any frontier model to explore the idea further. It includes your founder profile, the full venture context, all pre-research data (confidence, why now, personal moat, tools, risks), a locale-aware budget (<$500 USD, <$300-400 USD Asia, <¥3,000 CNY China), and an 8-section output structure: Go/No-Go verdict, sharpest niche, monetization, minimum lovable experience, lean architecture, market validation, 30-day sprint, risk modes, and 90-day dashboard. You can copy it to the clipboard or download it as markdown.

What happens if generation partially fails?

If one or two of the three idea slots time out or fail, you still receive whatever ideas completed successfully along with a visible banner indicating partial results. The credit is still deducted in this case because the pipeline produced usable output. Only a complete zero-idea failure is credit-free.

Do credits expire?

No. Your credits stay in your account indefinitely with no expiry date.

What counts as one credit?

Each "Generate" or "Load More" click uses exactly 1 credit, regardless of which model you choose or how long generation takes. The credit is deducted at the start of generation; if the pipeline fails completely and produces zero ideas, the credit is automatically refunded. Partial results (1 or 2 out of 3 ideas) do retain the credit deduction.

How much do credit packs cost?

There are three one-time credit packs, payable via Stripe: Small ($3 for 5 credits), Medium ($8 for 15 credits), and Large ($18 for 40 credits). There is no subscription — you only pay when you buy a pack.

Which models require credits?

All three models (DeepSeek R1, Qwen3-Max, Gemini 2.5 Pro) cost 1 credit per generation. The price per run is identical regardless of which model you select.

How do I get free credits?

New accounts receive 2 free credits automatically on sign-up (good for up to 6 ideas). No credit card is required — just create an account and start generating.

I paid but my credits did not arrive. What do I do?

Click "Restore Purchases" in the footer (signed-in users only). This scans your last 30 days of completed Stripe sessions and atomically adds any missing credits. The button is rate-limited to one call per five minutes per user. If credits still do not appear, contact support@smartmango.ai.

Why am I getting a "Too many requests" error?

To protect pipeline capacity, each IP address is limited to 10 generation requests per minute and a maximum of 3 concurrent running generations. Waiting a minute between bursts resolves most rate-limit errors.

Can I get a refund?

Purchases are generally non-refundable. However, if you experienced a technical issue that prevented generation from completing, contact us at support@smartmango.ai and we will review your case. If a charge is refunded or disputed, the corresponding credits are automatically removed (proportionally for partial refunds).

How does authentication work?

Authentication is handled by Clerk, a managed identity provider. You can sign up and sign in with email, social providers, or any method your organization configures in Clerk. We never store your password — only your Clerk user ID, which is used to look up your credit balance.

Is my profile information stored permanently?

No. Your profile inputs (skills, interests, location, time availability, pain points) are processed transiently during idea generation and are not permanently stored in our database. Your inputs are retained locally in your browser (localStorage) so the form can repopulate on return visits, but the server discards them after your generation run completes.

What data is stored in my account?

On the server we store only: your Clerk user ID, your Stripe customer ID (once you purchase credits), your credit balance, and your account creation timestamp. Generated ideas themselves are stored locally in your browser, not on our servers. Payment card details are handled securely by Stripe and never touch our infrastructure.

How is my input protected from misuse?

All user-submitted text passes through a multi-layer input sanitizer that strips 19 prompt-injection patterns (e.g., "ignore previous instructions", "[INST]", "system:") and control characters. Your inputs are then wrapped in <user_field> tags in the prompt, with explicit system-level rules forbidding the model from interpreting anything inside those tags as instructions.

How do I delete my account or data?

Contact us at privacy@smartmango.ai to request account deletion or a full data export. We will process your request in accordance with applicable data protection regulations including GDPR, CCPA, and PIPL.