AI-native tools, shipping on the App Store worldwide.
Each app below was designed around a single question: where can AI make a real task noticeably easier, faster, or more accurate? We don't ship novelty — every feature earns its place.
Wellness
Wellness coach
DailyKit puts a personal wellness coach in your pocket. Built-in Pomodoro timer, smart habit tracking with frequency goals, and a mood journal that spots patterns you might miss — all powered by an on-device LLM that adapts to your week.
Productivity
Kitchen assistant
ShelfSense knows what's in your fridge. Snap a shelf, log items manually, or scan a receipt — the AI tracks expiry dates, suggests recipes that use what you already have, and tells you what to buy before you run out.
Health & Fitness
Pet nutrition coach
PetPlate AI is your personal pet nutrition coach, grounded in AAFCO and AAHA public standards. Snap a photo of any pet food bag to score it against the official nutrient profile for your pet's species and life stage. Track weight, body condition, and feeding history per pet — built for the dog and cat owners who treat their pets as family.
Finance
Freelancer tax estimator
QuarterlyTax is an iOS app built specifically for the four estimated tax payments freelancers owe the IRS every year. Calculate your quarterly federal payment in under two minutes, track 1099 income and deductible expenses by category, and get reminders seven days before each deadline so you never pay an underpayment penalty.
More apps launching soon.
HOUJIAYAN Intelligence · Smart LLM Routing
Don't pick a model — let the platform pick for you. One OpenAI-compatible endpoint, one key for five providers and 16 models — no five separate signups. Smart routing matches each request to the best model automatically.
We aggregate five large-language-model providers behind a single, OpenAI-compatible API — but we don't stop at aggregation. The gateway reads your request, classifies its task type, and routes it to the model that will handle it best at the lowest cost. Set `"model": "auto"` and tell us your preference (cost, quality, or balanced); the gateway does the rest. Pay-as-you-go, no minimums, no enterprise procurement loop.
Walk into the AI-era equivalent of a triage desk — tell us what's wrong, we'll point you to the right specialist.
Walk into a hospital and you're told to pick your own department — you would, statistically, end up in the wrong one. Walk into one with a good triage desk, and you get to the right specialist on the first try. HOUJIAYAN's smart routing is the triage desk for your LLM calls.
Code, long-doc, translation, summary, writing, vision — the gateway reads your prompt, detects the task type, and dispatches to the candidate that leads that task's international authoritative benchmarks (primarily Artificial Analysis). Code goes to the strongest code model. Long documents go to the model with the largest context window.
Candidates are ranked by international public benchmarks. Only first-tier models within 5 points of the leader enter the quality band — anything weaker is never routed, no matter how cheap. Quality is decided before price.
Set `preference: quality` (default) to take the strongest in the band — the recommended default for the best result; `balanced` for the optional cost/quality/latency trade-off; `cost` to save wherever the band allows — cheapest in-band model wins, and at equal price the lower-latency, more stable channel wins. Your preference is honoured on every request.
Every 30 seconds the gateway probes each upstream. If one channel breaks, traffic switches to a healthy one and the task keeps running — rate-limited DeepSeek lands on Qwen, a Qwen latency spike shifts to MiniMax. Your code never has to handle a provider outage.
Every response carries a `routing_decision` field: which model was selected, which candidates were considered, why it won, and the benchmark source behind the ranking. A new model without international authoritative benchmark scores never enters smart routing — manual selection only; every strength judgement traces back to public international benchmarks. `model: "deepseek-v4-pro"` still works for explicit pinning.
curl https://api.houjiayan.com/v1/auto/chat/completions \
-H "Authorization: Bearer $HOUJIAYAN_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "auto",
"preference": "cost",
"messages": [{
"role": "user",
"content": "Write a Python function that validates an email address."
}],
"max_tokens": 500
}'{
"id": "chatcmpl-hjy-...",
"model": "deepseek-v4-flash",
"routing_decision": {
"selected": "deepseek-v4-flash",
"candidates_considered": ["qwen3.7-max", "doubao-seed-code"],
"reason": "code task detected, preference=cost, deepseek-v4-flash has lowest $/1M and code capability score 0.82",
"estimated_savings_vs_runner_up": "31%"
},
"choices": [{
"message": {
"role": "assistant",
"content": "def is_valid_email(email: str) -> bool:\n ..."
}
}]
}Notice: you didn't pick a model. You told the gateway what you wanted (a Python function, fast and cheap), and it picked `deepseek-v4-flash` for you. The `routing_decision` block tells you why and what it would have cost elsewhere.
We're not reselling keys. We're a technical-integration layer that routes your traffic to the best available provider.
One OpenAI-compatible endpoint, OpenAI/Anthropic/Gemini message formats. Drop-in for any client already talking to those APIs — change the base_url, keep the rest of your code.
One key behind five upstream providers and 16 models — no five platform signups, five contracts, or five billing relationships. Switch models with a single request field.
Per-token billing (input and output priced separately) at the upstream official list price; the payment-channel service fee is added on top at top-up — 5.0% crypto, 5.5% non-crypto (top up ¥100, receive ¥100, pay ¥105.00 / ¥105.50). No minimum top-up, no expiry, no per-seat fees. Detailed per-model rate sheet is published and updated monthly.
Cloudflare-fronted access from anywhere. Failover and per-key rate limits are handled in the gateway — your code stays the same whether the upstream is healthy or not.
Set "model": "auto" and the gateway classifies the task (code, long-doc, translation, summary, writing), then dispatches to the strongest in-band model on that task's international authoritative benchmarks — cheapest within the band, the more stable lane at equal price. Dual-channel failover: if one channel breaks, traffic switches automatically and the request doesn't fail.
We target 99% monthly availability at the public endpoint, measured on the public status page. Multi-upstream routing means individual provider outages rarely translate into user-visible downtime.
Mix-and-match — same endpoint, same request format. Or let smart routing pick for you.
Cost-effective, fast, strong at code and structured reasoning
→ Default for high-volume workloads. Use flash for chat, pro for harder reasoning.
Alibaba's flagship — strong on Chinese and multilingual tasks
→ Chinese-language agents, translation, code with Chinese-language comments.
Lightweight, multilingual, low-latency
→ Short-form generation, on-device-feel responsiveness, low-cost experiments.
ByteDance's model family — strong in Chinese, tuned for consumer AI
→ Chinese content, conversational tone, consumer-facing surfaces.
k3 flagship all-round with top-tier coding; k2.7-code series built for code; k2.6 high-value general
→ Agentic coding and long-context reasoning (k3, 1M context), dedicated code workloads (k2.7-code), cost-sensitive general tasks (k2.6).
OpenAI-compatible. Drop-in for any client written for the OpenAI API.
https://api.houjiayan.com/v1curl https://api.houjiayan.com/v1/chat/completions \
-H "Authorization: Bearer $HOUJIAYAN_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek-v4-flash",
"messages": [{"role": "user", "content": "Hello"}]
}'curl https://api.houjiayan.com/v1/models \
-H "Authorization: Bearer $HOUJIAYAN_KEY"Self-serve registration is live. Sign up, generate a key, and start calling the API in under a minute. New users get 100,000 free credits (about $1 worth).
Top up online in Console → Wallet: WeChat Pay or USDT-TRC20, with a payment-channel service fee (5.5% / 5%) added on top — credits land instantly after payment confirmation, ready to use immediately. New accounts get 100,000 free credits (about $1) on signup — no credit card required.
Pay-as-you-go. Prices are set in CNY (¥); US-dollar figures are approximate conversions at the weekly reference rate. Detailed rate sheet comes with your key.
| Model | Input / 1M | Output / 1M |
|---|---|---|
| DeepSeek v4-flash | $0.15 | $0.29 |
| DeepSeek v4-pro | $0.44 | $0.88 |
| Qwen3.7-Plus | $0.29 | $1.18 |
| MiniMax M3 (≤512K) | $0.30 | $1.20 |
| MiniMax M3 (512K–1M) | $0.60 | $2.40 |
| Volcengine Ark (Doubao) | $0.18 | $1.19 |
Models are billed per token (input and output priced separately) at the upstream official list price — we don't mark up tokens. At top-up, a payment-channel service fee is added on top: 5.0% for crypto, 5.5% for non-crypto payments. Top up ¥100 and exactly ¥100 lands in your balance — you pay ¥105.00 via crypto or ¥105.50 via non-crypto at checkout. The fee covers gateway operating costs (compute, tunnel bandwidth, and the time it takes to answer your support email). No hidden fees, no minimum top-up.
Same model family, two paths to reach it. Official list price (top-up fee added on top: 5.0% crypto / 5.5% non-crypto) vs. direct international pricing, July 2026.
| Model on HOUJIAYAN | In / Out ($/1M) | Comparable international | Direct price | You save |
|---|---|---|---|---|
| DeepSeek v4-flash | $0.15 / $0.29 | GPT-5.4-mini | $0.75 / $4.50 | 80% / 93% |
| DeepSeek v4-pro | $0.44 / $0.88 | GPT-5.4 | $2.50 / $15.00 | 82% / 94% |
| MiniMax M3 (≤512K tier) | $0.30 / $1.20 | Claude Sonnet 4.6 | $3.00 / $15.00 | 90% / 92% |
| Qwen3.7-Plus | $0.29 / $1.18 | Gemini 2.5 Flash | $0.30 / $2.50 | 1% / 52% |
We are not an authorised reseller of OpenAI, Anthropic, or Google. The international column is listed for reference only — the gateway itself does not proxy those providers. HOUJIAYAN aggregates five PRC-licensed model providers (DeepSeek, Alibaba, MiniMax, Volcengine, Moonshot) behind a single OpenAI-compatible interface.
Same model family, three ways to reach it. Free tier, our paid gateway, and the western incumbent — for developers picking where to send production traffic.
| Dimension | NVIDIA NIM (free) | HOUJIAYAN (paid) | OpenRouter (paid) |
|---|---|---|---|
| Price | $0 | Official list price + top-up fee added on top (5.0% crypto / 5.5% non-crypto) | Pass-through + markup |
| Rate limit | ~40 RPM, soft | 60 / 200 / 600 RPM by tier, hard | 1000 RPM (paid) |
| Models | ~140 (incl. PRC open-source) | DeepSeek · Qwen · MiniMax · Volcengine · Moonshot | ~200 (all major labs) |
| SLA | None — “dev and test” | 99% monthly target + public status page | Varies by tier |
| Invoice / receipt | No (free tier) | Yes | Yes |
| Human support | Forum | Email, by the engineer | Email (paid tiers) |
| Failover | None | Multi-upstream auto-retry | Per-provider |
| Mainland China access | Unreliable (GFW) | Cloudflare-fronted, stable | Unreliable |
NVIDIA NIM is a great free sandbox for a notebook or a weekend hackathon — and we’ll happily point you at it for that. When the API becomes a customer-facing surface, or you’d be embarrassed to explain a six-hour outage on a Monday morning, that’s when official-list pricing + a payment-channel top-up fee added on top (5.0% crypto / 5.5% non-crypto) starts paying for itself.
Verified on 2026-07-11 against build.nvidia.com
base_url.NVIDIA itself is upfront about this: the free tier is positioned for development and testing. The fine print is more important than the marketing:
A free API key is a fine place to start. A reliable one is what you ship.
The questions developers ask before they sign up.
Self-serve registration is live. Sign up, generate a key, and start calling the API in under a minute. New users get 100,000 free credits (about $1 worth).