Kagi Assistant is a private interface to major AI models. It’s run by Kagi, the company known for their private search engine.
With Kagi Assistant, your conversations are never used to train anyone’s AI. No account information is shared with model providers. Threads auto-delete after 24 hours of inactivity by default (adjustable in settings). Kagi publishes a full LLM Privacy Comparison so you can see exactly how each model handles your data.
The AI responses are integrated Kagi’s search engine, so it will honor any search customization you’ve created.
If you are asking sensitive medical or legal questions, you’ll probably want to opt for Maple AI. Maple encrypts your messages on your device before they ever leave it. It also has many other privacy protections that go beyond what Kagi offers.
Reviewers and privacy advocates suggest using them side by side for different tasks:
- Kagi Assistant → Quick research queries, search-backed answers, coding help, one-off factual questions. Its deep search integration makes it ideal for “look this up and summarize” tasks.
- Maple AI → Sensitive or personal conversations, long-running projects, document analysis, and anything you want truly private with persistent memory. Its end-to-end encryption and native apps make it better for ongoing, confidential work.
The Models
Kagi Assistant currently offers 34 models from 10 providers . Here’s the full list:
| Provider | Model | Plan |
|---|---|---|
| Alibaba | Qwen3-Coder, Qwen3.6 Plus | All |
| Anthropic | Claude 4.5 Haiku, Claude 4.6 Sonnet, Claude 4.8 Opus | Ultimate |
| Anthropic | Claude 4.5 Haiku / 4.6 Sonnet / 4.8 Opus (Reasoning) | Ultimate |
| DeepSeek | DeepSeek V4 Pro | Ultimate |
| DeepSeek | DeepSeek V4 Flash | All |
| Gemini 3.5 Flash, Gemini 3.1 Flash Lite (Preview) | All | |
| Gemini 2.5 Pro, Gemini 3.1 Pro (Preview) | Ultimate | |
| Mistral AI | Mistral Small 3, Mistral Small 4, Mistral Large 3 | All |
| Mistral AI | Mistral Medium 3.5 | Ultimate |
| MiniMax | MiniMax M2.7 | All |
| Moonshot AI | Kimi K2.5, Kimi K2.5 (Reasoning), Kimi K2.6, Kimi K2.6 (Reasoning) | All |
| Nous Research | Hermes-4-405B (Reasoning) | All |
| OpenAI | GPT 5.4 Mini, GPT 5.4 Nano, GPT OSS 120B | All |
| OpenAI | GPT 5.5, ChatGPT | Ultimate |
| xAI | Grok 4.3 | All |
| Z.ai | GLM-4.7, GLM-4.7 (Reasoning) | All |
| Z.ai | GLM-5.1, GLM-5.1 (Reasoning) | Ultimate |
Models marked “Reasoning” use extended chain-of-thought processing — better for complex logic, math, and multi-step problems.
You can switch models mid-conversation without starting over. If one model’s answer isn’t good enough, pick another and keep going.
Kagi Quick & Kagi Research: Agentic AI
Beyond simple chat, Kagi offers two “agentic assistants” that orchestrate multiple tools and sub-models automatically:
| Kagi Quick | Kagi Research | |
|---|---|---|
| Plan | All | Ultimate |
| Speed | <5 seconds | 20+ seconds |
| Focus | Direct, concise answers | Exhaustive, in-depth analysis |
Both assistants identify what to search for, run multiple simultaneous searches (including in different languages), and synthesize findings into sourced answers. Research Assistant can also run Python code, generate images, call Wolfram|Alpha, and perform news or location-specific searches.
Access them from the Assistant webapp or directly from Kagi’s search bar using bangs: !quick or !research.
See this blog post from Kagi for more information.
Pricing
Kagi Assistant uses a value-based usage system: your monthly plan determines your token allowance.
- Starter ($5/month): 300 searches + AI interactions; standard models only.
- Ultimate ($25/month): Unlimited searches + AI interactions; all models including premium ones like Claude Opus, GPT 5.5, and Kagi Research.
