Frequently Asked Questions

With discode you get access to the leading European AI chat platform that automatically routes your requests to the best AI model from 100+ models — including GPT, Claude, Gemini, Llama, DeepSeek and Mistral.

ChatGPT only uses OpenAI models. discode gives you access to 100+ models from 12 providers and automatically selects the best one for your query. You pay one subscription instead of five — and get data processing in the EU with anonymization of sensitive data.

100+ models: GPT-5 Mini, GPT-4o Mini, Claude Sonnet 4.6, Gemini 2.5 Pro, Gemini 2.5 Flash, Llama 3.3 70B, DeepSeek V3, DeepSeek R1, Mistral Large, Perplexity Sonar and more — all connected via OpenRouter.Learn more →

Free: €0 (20 queries/day, Tier 1–2). Starter: €9.99/month (300 queries/month, all tiers 1-4). Pro: €17.99/month (800 queries/month, Trio, Deep Dive, Challenger). Early Bird offers start at €1.

Yes! The Free plan offers 20 queries per day with access to Tier 1–2 models, Autoplay mode and Eco Score. No credit card needed.

Solo is the default routing mode: when you don't select Trio, discode's routing engine automatically picks the optimal model from 100+ models. The system categorises your request (math, code, creative etc.) and routes to the best model in the 4-tier pool — from fast-cheap (Tier 1) to Frontier (Tier 4).Learn more →

The Finetuning menu (funnel icon on the right of the input field) gives you three controls: Smarter (stronger models, deeper verification), Speed (faster response) and Eco (resource-efficient data centres). In default mode they are on/off toggles. In Expert Mode (toggle top-right) you can fine-tune each from 1–5.Learn more →

The three Finetuning filters control model routing: Smarter prioritises qualitatively stronger models for more precise answers. Speed selects faster, cheaper models for lower latency. Eco routes to data centres with lower CO₂ emissions. All three can be active simultaneously. In Expert Mode each is adjustable from 0–5.Learn more →

Trio has 3 AI models from different manufacturer families respond simultaneously — e.g. GPT-5 Mini + Claude Sonnet 4.5 + Gemini 2.5 Flash. Different families have different error patterns and correct each other. A judge evaluates all three answers blindly, a challenger questions, a synthesis wraps up. Research shows: hallucination rate drops from ~5–25 % to under 2 %.Learn more →

Challenger runs in 3 rounds: Round 1 (Critic) analyses the original answer and identifies gaps and unsupported claims. Round 2 (Improver) builds in the corrections. Round 3 (Judge) evaluates and synthesises the result. Research shows that LLMs cannot self-correct — a separate Challenger from a different model family is essential.Learn more →

Solo: one model answers automatically — fast, cost-efficient, ideal for direct questions. Trio: three models from different families answer in parallel, a judge evaluates, a challenger questions, a synthesis wraps up — hallucination rate under 2%. Solo for everyday use, Trio for topics where precision and fact-checking matter.Learn more →

The Eco Score shows the estimated ecological footprint of each request: CO₂ (g), water consumption (ml) and energy demand (Wh) — per message, displayed directly below every response. The Eco slider in Mixer Mode prioritises more resource-efficient models and EU providers.Learn more →

The Eco Score (CO₂ and water consumption) is an estimate [Beta]. We calculate the approximate consumption based on the number of processed tokens and published averages for energy consumption of various AI models. These values are NOT precisely measured. Actual consumption depends on many factors: provider hardware, server load, energy mix at the location. We therefore label all eco values with [Beta]. Our goal: create transparency and raise awareness of the ecological footprint of AI usage — not provide an exact accounting.

In Mixer mode, you can select a specific model via the manual model chip. In Autoplay mode, the routing system chooses automatically.

Privacy is built into discode from the start: personal data (PII) is anonymized locally before it's sent to AI providers. Our servers are located in Frankfurt (EU), and we don't store any data on the AI providers' servers.Learn more →

All data is stored in the EU: Supabase (PostgreSQL) for chat data, Clerk for authentication, Vercel EU region (Frankfurt) for the application.

Your chats are stored encrypted in your personal account and protected with Row-Level Security (RLS). Only you have access to your data.

No. discode does not store data on AI provider servers. Your requests are forwarded anonymized and not used for training.

PII (Personally Identifiable Information) anonymization automatically detects sensitive data in your messages and replaces it with placeholders before the request is sent to AI providers. The feature is optional and can be activated per request or globally.

Automatically detected and anonymized (with active anonymization chip): Email addresses · Phone numbers (AT/DE/international) · IBAN numbers (mathematically verified) · Credit card numbers (Luhn-verified) · IP addresses · Postal code + city · Tax identification numbers · VAT IDs · Insurance numbers · Account numbers · Dates of birth · Street names with house numbers · Personal names (via AI detection) · Company names (via AI detection). ~100 tech/company names are automatically ignored to prevent false positives.

Uploaded PDFs and documents: Text is extracted and anonymized. Embedded images in PDFs are NOT analyzed. — Images with text (e.g. photo of an ID): Are not processed by anonymization. Please don't upload images with sensitive data. — Indirect identifiability ("the mayor of [small town]"): Is not detected. — Non-Latin characters: Limited detection. — The anonymization detects patterns, not context. When in doubt: remove sensitive data manually.

In case of technical failure, your message is NOT sent (Fail-Closed). You receive an error message. Your data does not leave the server in this case.

Your requests are forwarded via OpenRouter to AI models that partially run on US servers (e.g. OpenAI, Anthropic, Google). Only the message text is transmitted — no account data, no IP address, no metadata. With an active GDPR chip, personal data (names, emails, phone numbers etc.) is automatically replaced by placeholders beforehand. The response is re-identified and stored only in your EU account. AI providers only receive the anonymized message text via the API. According to their API terms, API requests are not used for training. See the individual providers' privacy policies for details.

Without an active GDPR chip, your message text is sent unchanged to the AI provider. Personal data like names, email addresses or phone numbers will then NOT be anonymized. The decision is yours — you can activate or deactivate the GDPR chip at any time with a click. For sensitive content, we recommend always keeping anonymization enabled.

Early Bird is a limited-time offer for early users: 3 months Pro plan from just €1. Spots are limited and staggered in tiers.

20 queries per day, access to Tier 1–2 models (Llama, Gemini Flash, Mistral), Autoplay mode and Eco Score. No credit card needed, no time limit.

Founder's Pricing are discounted rates for early users. Current prices (Starter €9.99, Pro €17.99) are introductory prices that remain for existing customers.

Yes. All subscriptions are cancellable monthly. After cancellation, you keep access until the end of the billing period, then you automatically switch to the Free plan.

We accept credit cards (Visa, Mastercard, AMEX), SEPA direct debit and other local payment methods through our payment provider Stripe.

Transparency is a core principle of discode. You see a cost estimate before every request and the actual deducted amount in cents after every response. So you always know what your AI usage costs — no model is a black box. Many providers hide actual costs behind flat subscriptions. We show them because we believe informed users make better decisions.

Yes, discode is a Progressive Web App (PWA), optimized for mobile devices. You can also add it to your home screen for an app-like experience.

discode is a PWA (Progressive Web App) and can be added to the home screen on iOS and Android. A native app is currently not planned.

Yes, discode supports image generation via AI models and video generation (experimental). Image uploads for vision models are planned.

The maximum message length depends on the model used, but is typically 4,000–8,000 characters per message.

The interface is available in 6 languages: German, English, French, Spanish, Italian and Portuguese. The AI models themselves support over 100 languages — you can chat in any language.

discode was founded by Moriz Piffl-Percevic, with a small team of UX design and system architecture. We are a Vienna-based startup with the goal of democratizing AI access in Europe.

discode is based in Vienna, Austria. Our servers are in Frankfurt, Germany (EU). We are pro-European — not anti-American.