About — AI Olympiad

About AI Olympiad

AI Olympiad goes beyond standard side-by-side model comparison. It is a local-first, Bring Your Own Key (BYOK) platform that lets you compare top AI models, route prompts intelligently, and objectively judge the best outputs. With AI Olympiad, your keys, your data, and your costs remain entirely in your control.

View plans Enter Events

The Olympic Metaphor

We use a sports metaphor to make managing multiple AI models intuitive and fun:

Event: A chat session.
Athlete: An AI model.
Lanes: Side-by-side model generations.
Podium + Judge's Citation: The final ranking and evaluation of the models' answers.
Coach: The intelligent model picker.
Vault: Where your API keys are stored safely.

Core Features

The Podium & Judge

In Podium mode, a single prompt fans out to up to six different models (Lanes). Once all models finish generating their responses, an impartial Judge ranks them 1st, 2nd, and 3rd (Gold, Silver, and Bronze) and provides a one-sentence "Judge's Citation" explaining the decision.

Impartiality: The Judge is never one of the competitors. It uses a cost-efficient model from your vault (typically Gemini Flash, GPT-5 Mini, or Claude Haiku).
Scoring Criteria: Models are judged strictly on accuracy, instruction-following, formatting, and completeness. Brand prestige is ignored.

The Coach (Pro / Lifetime)

The Coach is your cost-aware model router. It analyzes your prompt while you type and recommends the cheapest "athlete" capable of winning the event—saving you money without sacrificing quality.

Heuristic (Instant, Local): Detects the prompt's intent (code, math, creative, translation, etc.) and complexity, then selects the best option from your available keys.
Classifier (Optional): Uses a highly affordable model to confirm the pick in JSON, falling back to the heuristic if needed.

Live Metrics

Every finished lane features a live scoreboard displaying:

Time: Wall-clock generation time from the first token to completion.
Tok/s: Output tokens per second (stream speed).
Cost: Estimated USD based on token usage multiplied by the provider's published per-million rates. (Your provider bills you directly, not AI Olympiad.)

Your Keys, Your Data

AI Olympiad is built on a privacy-first architecture.

Your API keys live exclusively in your device's Vault (with optional AES-GCM encryption).
Your chat history is stored locally in IndexedDB and is never synced to the cloud or a third-party account.
You can install AI Olympiad as a Progressive Web App (PWA) directly from your browser's "Add to Home Screen" prompt.
See Lane Congestion info below.

Plans & Pricing

AI Olympiad is a single-user subscription tied to your email. We do not sell model credits; you pay your chosen AI providers (OpenAI, Anthropic, Google, Groq, etc.) directly.

Free Trial: Sign in with your email for a short, full Pro trial before subscribing.
Single ($5/mo or $49/yr): Single-mode chat + Vault UI.
Compare ($7/mo or $75/yr): Compare mode + Podium (2 lanes + Judge).
Pro ($9/mo or $99/yr): All modes + Coach + 6-lane Podium.

View plans & checkout →

Get Started in Five Minutes

Sign In and Choose a Plan: Free trial available. Log in using your email address to activate your free Pro trial, or select the plan that fits your needs.
Enter the Locker Room: Navigate to the Locker Room on the Events screen to manage your local API Vault.
Add Your First Provider: Only one key is required to start. Choose Google, Groq, or OpenAI and paste your API key.
Test and Save: Tap Test key, then save to your Vault. Optional AES-GCM encryption available.
Enter the Arena: Start a new Event. On Pro, use Coach recommendations or switch to Podium mode.

Frequently Asked Questions (FAQ)

Do I need an API key for every single provider on day one?: No. A single key is enough to get started. If you add one provider, you can immediately begin comparing different models offered by that specific provider.
Which AI acts as the Judge?: The system selects a cheap, judge-eligible model from your own Vault (like Gemini Flash or GPT-4o Mini). The Judge will never be one of the competitors in that specific Podium race.
What exactly does the Coach do?: The Coach classifies your prompt's intent (e.g., coding, creative writing, math) and complexity, then recommends the most cost-effective model in your Vault capable of handling that task.
How do the metrics and cost estimates work?: The scoreboard tracks real-time performance: wall-clock time and tokens per second. Cost is estimated from token usage × the provider's published API rates.
Where is my chat history and data stored?: Everything stays on your device. API keys are kept in your local Vault; chat logs are saved to IndexedDB. We never sync your data to a cloud account.
What does the "Lane Congestion" message mean?: Lane Congestion means an athlete's stream is being throttled or delayed—often from provider rate limits (HTTP 429), browser connection limits, or running many lanes from the same provider at once.

Understanding "Lane Congestion"

When running multi-model events in Compare or Podium mode, you may occasionally see a Lane Congestion warning on one or more lanes. Because AI Olympiad runs client-side and fans out prompts to multiple AI APIs simultaneously, congestion occurs when a lane experiences throughput bottlenecks, stream delays, or API rate throttling.

Why Lane Congestion Occurs

Cause	What's Happening
Provider Rate Limits (RPM / TPM)	Multiple lanes from the same provider may hit Requests-Per-Minute or Tokens-Per-Minute caps, triggering HTTP 429 Too Many Requests.
Browser Connection Limits	Browsers cap concurrent HTTP/2 and SSE streams per domain (typically 6). Six heavy streams can saturate network sockets.
Main-Thread Rendering Load	Hundreds of tokens per second across multiple streams, written to IndexedDB, can briefly overload browser rendering.
Provider Queue Latency	High-demand frontier models may experience server-side queue delays during peak usage hours.

How to Prevent and Resolve Lane Congestion

Diversify Your Athlete Roster (Multi-Provider): Mix providers across lanes (e.g., Google, Anthropic, OpenAI, Groq) to distribute API limits.
Check Your Provider Account Tiers: Free-tier keys have strict concurrency caps; upgrading increases TPM/RPM thresholds.
Reduce Concurrent Lanes: On mobile or low bandwidth, scale back from 6 lanes to 2–3.
Lower Output Token Length: Shorter max output keeps streams open for less time.
Enable Auto-Backoff / Staggered Start: When available in settings, staggered fan-out sends prompts in micro-intervals rather than all at once.