Research
Building intelligence that understands you
Explore Torki's AI models, capabilities, safety practices, and the technology powering the next generation of personal AI.
Models
Our models
Four purpose-built models, each optimized for different tasks — from quick answers to deep reasoning and complex multi-step work.
Pulse is Torki's fastest model, optimized for everyday tasks that need quick, accurate responses. It excels at Q&A, brainstorming, summarization, and rapid-fire conversations where speed matters more than depth.
Best for
Available on Free, Lite, Arc, Zenith
Orbit is Torki's general-purpose model, striking the ideal balance between speed and intelligence. It handles writing, analysis, research, and complex conversations with consistent quality and reliability.
Best for
Available on Free, Lite, Arc, Zenith
Stiller is Torki's deep reasoning model, designed for problems that require extended thinking. It takes more time but produces thorough, step-by-step analysis for math, logic, multi-step problems, and nuanced research.
Best for
Available on Lite, Arc, Zenith
Cosmos is Torki's most capable model, built for the most demanding tasks. It handles complex code generation, agentic workflows, long-form creative writing, and multi-step tasks that require maximum intelligence and context utilization.
Best for
Available on Lite, Arc, Zenith
Comparison
Compare models
| Capability | Pulse | Orbit | Stiller | Cosmos |
|---|---|---|---|---|
| Response speed | ||||
| Reasoning depth | ||||
| Code generation | ||||
| Creative writing | ||||
| Math & logic | ||||
| Conversation | ||||
| Summarization | ||||
| Context utilization | ||||
| Multi-step tasks | ||||
| Output length | Standard | Standard | Higher | Highest |
| Context window | 200k | 200k | 200k | 200k |
Platform
Beyond models — a complete AI platform
Torki isn't just a chatbot. It's a full AI platform with six integrated capabilities.
Chat
Multi-turn conversations with memory, context, and 7 unique AI personalities — each a specialist in different domains.
Voice (RTV)
Real-time voice conversations powered by Torki's RTV engine. Talk naturally with any AI personality — on web, iOS, and Android.
Vision
Generate, edit, and transform images with Kvora — Torki's image generation system. Text-to-image, image-to-image, and upscaling.
Canvas
Write, run, and preview code in an interactive sandbox. Supports multiple languages with live output and iteration.
Agents
Build, publish, and embed custom AI agents with specialized knowledge, tools, and personality. Connect via MCP.
Research
Deep research mode with multi-query decomposition, extended analysis, and cited sources for thorough investigation.
Safety
Our approach to safety
Building powerful AI means building responsible AI. Here's how we think about safety at Torki.
Data Privacy
Your conversations are yours. Torki does not use your data to train models by default. All users have the option to opt out of any data sharing. We process data only to deliver responses and improve your experience — never to sell or share with third parties.
Content Safety
Torki's models include built-in safety measures to prevent harmful, misleading, or inappropriate outputs. Our safety systems are continuously updated to address new challenges while maintaining the helpfulness and creativity that makes Torki useful.
Transparency
We believe in being honest about what AI can and cannot do. Torki's research mode cites sources. Our models acknowledge uncertainty. We're building AI that's helpful without being deceptive.
Security
All data is encrypted in transit and at rest. Torki uses industry-standard security practices including bearer token authentication for API access, secure OAuth for social logins, and regular security audits.
Technology
Under the hood
200k context window
Process up to 200,000 tokens of context across all models. Upload documents, have long conversations, analyze large codebases.
Multi-modal input
Text, images, documents, code. Torki understands all of them natively with seamless processing across input types.
Real-time voice
Sub-second latency voice conversations powered by Torki's custom RTV (Real-Time Voice) pipeline across all platforms.
Agent framework
Build agents with custom system prompts, knowledge bases, tool connections, and MCP integration. Publish and embed anywhere.
7 AI companions
Torki, Elli, Poppy, Eywa, Kalli, Pooline, and 61 — each with distinct personality, voice, and domain expertise.
Cross-platform
Available on web (torkiai.com), iOS (App Store), and Android (Play Store) with synced conversations across devices.
Updates
Research updates
The latest from our research and engineering teams.
Introducing Torki's Model Lineup: Pulse, Orbit, Stiller & Cosmos
A deep dive into Torki's four AI models — their architecture, capabilities, and how they work together.
Coming soon
Building Real-Time Voice for AI: The RTV Engine
How we built sub-second voice conversations across web and mobile platforms.
Coming soon
Torki's Approach to AI Safety and Privacy
Our principles, practices, and commitments for building responsible AI.
Coming soon