Models & limits
The short version. Both Lite and Pro include the Keyda AI cloud pool — no API key needed. Lite gets 200 requests per day with up to 4,096 output tokens per response; Pro gets 1,000 requests per day, a 500,000-token daily cap, and up to 2,048 output tokens per response. We rotate models as providers ship better options so you do not have to manage the line-up yourself.
1. What's included with Lite & Pro
Both Lite and Pro proxy your AI calls through our infrastructure using our keys, so you never have to manage your own API key. We support the most capable model from each of the leading AI labs and keep them upgraded as new versions launch. The only difference is the daily request quota: 200/day on Lite, 1,000/day on Pro.
2. Current cloud models
These are the cloud models live today. The list is updated whenever providers ship a new version — Lite and Pro users get the upgrade automatically at no extra cost.
| Provider | Models | Best for |
|---|---|---|
| OpenAI | GPT 5.5, GPT 5.4 Pro, GPT 5.4 Mini / Nano, GPT 4.1, GPT 4.1 Mini | Reasoning, multimodal, daily-driver writing |
| Anthropic | Claude Opus 4.7, Claude Sonnet 4.6, Claude Haiku 4.5 | Long-context, code, balanced writing, fastest replies |
| Gemini 3.1 Pro, Gemini 3.1 Flash Lite, Gemini 2.5 Pro / Flash / Flash Lite | Multimodal, price-performance, fastest inference | |
| xAI | Grok 4, Grok 4 Fast (reasoning / non-reasoning), Grok Code Fast | Real-time knowledge, fast reasoning, agentic coding |
| Groq | Llama 3.3 70B, Llama 3.1 8B Instant, GPT-OSS 120B / 20B (via Groq) | Ultra-low-latency open-source inference |
Model availability can change without notice if a provider deprecates a model or releases a successor. We keep the line-up state-of-the-art — Lite and Pro users do not need to track this.
3. On-device models (Lite & Pro)
These models generate text on your phone: once a model is downloaded (from Hugging Face), running it needs no network, costs no tokens and has no rate limit. Assistant chats made with an on-device model are still saved to your account like any other chat. They are smaller than cloud models but excellent for short replies, grammar fixes, summaries and quick rephrases.
- Gemma 3n (best quality) — Google's on-device flagship; recommended for most users.
- Gemma 3 1B (fast) — smaller and faster; great for older devices.
- Additional optimised open-source models added as the on-device ecosystem matures.
On-device image generation is also included on Lite and Pro for sticker creation — fully offline.
4. Daily limits
| Limit | Lite | Pro | Notes |
|---|---|---|---|
| Cloud AI requests / day | 200 | 1,000 | Resets at 00:00 IST |
| Tokens per day | — | 500,000 | Sum of input + output tokens across all cloud calls |
| Tokens per single response | 4,096 | 2,048 | Per-request output ceiling |
| On-device usage | Unlimited | Unlimited | Generated on your phone — no quota |
| BYOK usage (own key) | Unlimited | Unlimited | Bound only by your provider's own limits |
Quotas reset daily at 00:00 IST. Once the daily request cap is reached, cloud AI calls are paused until midnight — on-device models and BYOK remain usable without restriction. If you regularly hit your cap, upgrading to Pro gives 5× more requests per day.
5. How model selection works
By default Keyda picks a sensible model per provider for you (the current "daily driver" for that lab). You can override the default per provider from Settings → AI → Model inside the keyboard if you want a specific model. If a chosen model fails or is deprecated, Keyda falls back to the next-best model from the same provider automatically.
6. Privacy
Keyda does not train AI models on your prompts or responses; how each AI provider handles the requests it receives is governed by that provider's own terms. To power your in-app usage stats and to detect abuse, we keep a log of the AI requests you make through Keyda's AI — including the prompt and the result, plus the model, a token estimate and timestamp — tied to your account; by default the same is recorded for requests made with your own key (see the Privacy Policy, section 2.4). These logs are kept for about 12 months and then deleted. Grammar-fix and rephrase requests also pass through a short-lived answer cache shared across users, described in the Privacy Policy, section 2.5, which has the full detail.
7. Changes
We may add, swap or retire models at any time as the AI landscape evolves. Material changes (such as a daily-cap reduction) will be announced in-app and on this page. The "Last updated" date at the top reflects the current version.
8. Contact
Questions about which model to use, or hit a limit? Email support@keyda.in or use the Contact form.