Everything you need to know about building with Nataris.
Your prompts are transmitted securely (TLS/HTTPS) to our backend, then routed to a provider device for processing. During inference, your prompt is visible to the provider device. Unlike centralized clouds, inference happens on distributed devices. We store the prompt and response with the job record for billing, support and disputes — never for training, and never for sale. If you opt into conversation memory, that data is stored server-side at your request and can be deleted anytime.
For more details, see our Security Architecture.
Great for: General AI tasks, chatbots, content generation, creative writing, code assistance, prototyping, and any use case where you want decentralized inference without vendor-controlled data pipelines.
Not recommended for: Personal Identifiable Information (PII), protected health information (PHI), financial data, trade secrets, or anything requiring compliance certifications (HIPAA, SOC2).
Yes, with the right use cases. Nataris is production-ready for general AI tasks, public-facing content, and applications that don't handle sensitive personal data. Many developers use us for chatbots, content generation, and creative tools.
Jobs automatically retry on a different provider (up to 3 attempts). You only pay if the job completes successfully. Average retry time is under 2 seconds. Our smart routing prioritizes reliable providers based on their track record.
Standard routing uses all available providers. Verified routing (set privacyLevel: "verified") routes only to providers whose devices have passed Google Play Integrity verification — confirming the device and app haven't been tampered with. This reduces risk from modified or rooted devices.
Sign up for an account, get your API key, and start making requests. Our API is OpenAI-compatible — if you've used OpenAI, you can switch to Nataris by changing two lines of code (API key and base URL).
The /v1/chat/completions endpoint is our OpenAI-compatible chat API. It accepts a messages array (with roles like "user", "assistant", "system") instead of a single prompt string. It also supports optional orchestration and conversation_id fields for advanced features.
Context management: Each request may hit a different provider device. You can either send full conversation history each time, or use our server-side conversation memory — pass a conversation_id and the API will automatically persist and recall message context.
Orchestration lets you run complex, multi-step AI tasks through a single API call. Instead of making multiple requests yourself, the backend chains inference steps automatically. Available workflow types:
Add "orchestration": {"enabled": true, "workflow": "research"} to your chat completions request. Orchestrated steps are billed at the same base model rate — no surcharge. You can set a budget cap via max_cost_usd.
Yes. Nataris supports server-side conversation memory. Create a conversation via POST /v1/conversations, then pass the conversation_id with each chat request. The API will automatically persist messages, generate a title from the first message, and create rolling summaries of older messages to maintain context efficiently.
You can also manage conversations without memory — just send the full message history with each request as with any OpenAI-compatible API.
Yes. We apply generous but reasonable limits to keep the platform sustainable:
You can always delete your data manually via the API at any time. The auto-cleanup is especially useful in LibreChat-style setups where many transient conversations accumulate over time. Active data is never touched.
Nataris is OpenAI-compatible, so it works as a drop-in backend for LibreChat, Open WebUI, and similar frontends. Set your API Base URL to your Nataris endpoint and use an API key for authentication.
Since these frontends send standard OpenAI requests (without Nataris-specific fields), you can configure per-API-key feature defaults in the Developer Portal under API Keys → Configure. Enable Auto Orchestration or Auto Conversation Memory, and every request made with that key will automatically get those features without any changes to the frontend.
For conversation memory, sessions are detected automatically based on inactivity timeout (default 30 minutes). You can also send an X-Session-ID header for explicit session control.
Yes. Every new developer account receives $5 (≈₹425) in free API credits — no payment required. This is enough for thousands of inference calls, depending on the model. You can add more credits via Razorpay (UPI, cards, netbanking) — payments are processed in INR.
We use a prepaid credit system. You add credits to your account and pay per inference based on tokens processed. Pricing varies by model — smaller models are cheaper. You'll never be charged more than your available balance. View your usage and balance in the dashboard.
Nataris runs inference on real devices — mostly smartphones — so response times are measured in seconds, not milliseconds. For small models (0.5B–1B), expect 2-5 seconds. For Phi-3 Mini (3.8B), expect 15-20 seconds. This is a trade-off of decentralized, people-powered infrastructure.
If a device fails mid-job, smart failover retries on the next available device within seconds. Best for use cases where freedom and flexibility matter more than sub-second latency.
We serve open-weight models (Llama, Mistral, Phi, Qwen) as-is, without adding vendor-mandated content moderation or output filtering. The models produce whatever they were trained to produce. We don't intercept, censor, or modify outputs.
This is especially useful for creative applications, fiction writing, research tools, and any use case where vendor content policies would restrict your work. You remain responsible for how you use the outputs in your application — see our Acceptable Use Policy.
Some carriers may delay or filter OTP messages. If you don't receive the code within 60 seconds, try the "Resend OTP" button. You can also sign in with email — just tap "Sign in with email instead" on the OTP screen. Both methods give you full access.
No. We have multiple safeguards: auto-pause when battery drops below 20%, auto-pause when temperature exceeds safe thresholds, and optional "only while charging" mode. You can also set scheduled hours (e.g., overnight only) to minimize impact on your daily use.
Some phones restrict background apps to save battery. To stay online and receive jobs: open your phone's Settings → Apps → Nataris → Battery, and set to Unrestricted (or "Don't optimize"). That lets Nataris run in the background so you keep earning.
You earn for each inference job your device completes. Earnings depend on the model size (larger models pay more), your device specs (faster devices get more jobs), network demand, and your uptime. Withdrawals are available once you reach the minimum: ₹50 (India) / $5 (international).
Detailed earnings estimates coming soon. Actual earnings vary significantly based on these factors.
Go to Settings → Payout Settings in the app. Add your UPI ID for India payouts, or bank details where available. In India, PAN may be required for larger payouts. Request a withdrawal when your balance exceeds the minimum. Payouts are processed within 3–5 business days after approval.
Yes. One account, unlimited devices. Each device earns independently. Many providers run old phones they're not using as dedicated earning devices.
As a provider, you agree not to store, log, or share any prompts or responses that pass through your device. Violations result in account termination and may have legal consequences. You'll confirm this agreement during onboarding.
Most people are asking one of these
The full answers are below, in three short groups.
Optional paid surveys from CPX Research GmbH, a market research company in Germany. Brands and research institutes pay for people's opinions. You can use the app and earn from AI jobs without ever opening them.
Open the app and tap Earn. If the tab is not there yet, close the app fully and open it again. It appears within about fifteen minutes of us switching it on, or as soon as you reopen the app.
Usually a few minutes each. The pay is shown on each survey before you start. We do not publish an average because we do not yet have enough real completions to give you an honest one.
800 coins is one US dollar. The app converts coins into money in your balance automatically. You never need to do the sum yourself.
You were screened out. Surveys look for particular kinds of people, and partway through, one may decide you are not who it needs. It is the most common thing that happens on any survey wall, and it is not a problem with your account.
The survey company does not pay us for a screen-out, so there is nothing for us to pass on. Occasionally one pays a small amount for your time, and when it does, you get it. Just pick another survey.
Which surveys you see depends on your profile and your country, and it changes through the day. Check back later, especially in the evening.
It shows as pending straight away. It moves into your balance in the first week of the month after you did the survey. That is when the survey company confirms the month and pays us. It happens on its own.
AI pay comes straight from us, so it lands immediately. Survey pay comes from the survey company, which confirms each month before paying anyone. Both are real and both are yours.
Yes, if the survey company rejects that completion. Normally that happens before it confirms the month, which is why the money waits, and the amount is simply removed from pending. In rare cases it happens after, and if the money is still in your balance it comes back out. It is the survey company's decision, not ours.
Once it reaches your balance it is the same as any other earning. The same minimum and the same payout options apply. See the withdrawal answer above.
The survey company asks profile questions such as age, gender and postcode to match you to surveys. We are only allowed to pay adults for that. You confirm it once and are not asked again.
It is recorded and held for you, not lost. Update the app from the Play Store, confirm your age when asked, and the money joins your balance on the normal schedule.
We share only your Nataris account ID. Never your name, email or phone number. We do not see your answers. We only learn that a survey was finished and how much it paid. Your answers are handled by CPX under their own privacy policy, and our privacy policy covers the rest.
We currently support 3 text/code generation models optimized for mobile P2P inference:
Akash and Render use datacenter GPUs — high upfront costs and complex setup. Nataris uses mobile phones — low barrier to entry, 2+ billion potential provider devices worldwide. Our unique advantage: Google Play Integrity verification for trust, which isn't possible with traditional compute networks.
Not yet. Token launch is planned for a later phase (see our Roadmap). For now, you pay with credit card (Stripe) and providers receive payouts in local currency. Future token benefits will include lower transaction fees, staking rewards, and governance voting rights.
Developers pay per token. Providers keep 80% of each job they complete — 80% is yours. See Pricing for per-model rates.
Check our documentation or reach out to our team.
One message is enough. We are a small team and we read everything. Sending several in a row does not get you a faster answer, it just means the same question is read several times before it is answered once.