Features

100% On-Device AI

Every model runs entirely on your iPhone or Android device using hardware-accelerated inference. No API key. No Wi-Fi. No cellular. When airplane mode is on, Zima still works — because there is nothing to connect to.

Three Model Sizes

Choose the right model for your device and use case:

• Nano (559 MB, Free) — Ultra-fast for quick questions, brainstorming, and everyday tasks. Fits every modern phone.
• Plus (2.7 GB) — Balanced power and speed. Richer reasoning, longer context, better writing.
• Pro (4.7 GB) — The most capable on-device model available. Best for complex analysis, code review, and nuanced conversation.

All models download once and run offline forever. No subscription required for Nano.

Streaming Responses

Words appear word-by-word as the model generates them — exactly like a top-tier cloud AI. Tap the stop button to halt generation mid-stream, or use Regenerate to get a different response without losing your conversation history.

Eight Personalities + Custom

Zima ships with eight distinct personalities you can switch between instantly:

• Helpful — Clear and direct. The default.
• The Butler — Formal, precise, impeccably polite.
• The Pessimist — Technically accurate. Quietly skeptical.
• The Intern — Eager, literal, always trying.
• The Detective — Every answer is a deduction.
• The Bard — Poetic, metaphorical, vivid.
• The Theorist — First principles. Second-order effects.
• The Surfer — Chill. Vibes-first. Still knowledgeable.

Or write a fully custom system prompt to define your own AI persona.

Markdown & Code Rendering

Responses render with full markdown support — headers, bold, bullet lists, numbered lists, and syntax-labeled code blocks. Ask for code and get back formatted, readable output, not a wall of monospace text.

Conversation History with Search

All conversations are saved locally on your device. Full-text search lets you find anything you discussed — no cloud account required, no sync, no privacy tradeoff.

Export & Share

Long-press any message to copy it. Share individual responses with the system share sheet. Export an entire conversation in one tap to save or send anywhere.

Context Limit Indicator

A visual bar shows how much of the model's context window is in use. At 95% capacity, Zima blocks new input and lets you start a fresh conversation — no silent truncation, no unexpected behavior.

Degenerate Repetition Detection

On-device models can occasionally fall into looping patterns. Zima detects this automatically and stops the generation before it becomes a problem.

Haptic Feedback

Subtle haptic responses throughout the interface — on send, on message copy, on generation complete. The app feels alive without being intrusive.

No Account. No Tracking. No Subscription (Nano).

Download the app, open it, and start chatting. No email. No password. No onboarding screen asking for permissions it doesn't need. The Nano model is permanently free. Plus and Pro are one-time in-app purchases — no recurring charges.