Features
Built for people who already know which model is best at what.
The same chat surface across every frontier provider, plus the small things that make daily use bearable: branching, side-by-side compare, keyboard shortcuts, and a Canvas pane for everything models render.
| Provider | Models |
|---|---|
| OpenAI | GPT-5, GPT-4.1, o3 |
| Anthropic | Claude Opus 4, Sonnet 4 |
| Gemini 2.5 Pro, Flash | |
| Mistral | Large, Codestral |
| DeepSeek | V3, R1 Reasoner |
| xAI | Grok 3, Grok 3 Mini |
| OpenRouter | 300+ models |
| Tavily | Web search |
Models
Every frontier provider in one composer.
We integrate directly with eight providers and pass through their native streaming, tool use, and reasoning APIs. No homegrown wrappers that fall behind the upstream — when Anthropic ships extended thinking, you get extended thinking the same day.
- OpenAI: GPT-5, GPT-4.1, o3, o4-mini reasoning
- Anthropic: Claude Opus 4, Sonnet 4, Haiku with extended thinking
- Google: Gemini 2.5 Pro and Flash with native multimodal
- Mistral: Large, Codestral, Nemo
- DeepSeek: V3 chat and R1 reasoner
- xAI: Grok 3 and Grok 3 Mini
- OpenRouter: 300+ additional models, including open weights
- FAL: image generation and edit models inline
Chat experience
Designed for people who actually use chat all day.
Branching, regeneration with a different model, side-by-side compare, keyboard-first navigation, drag-to-attach files. The interface is dark by default and stays out of your way.
- Branch any message to try a different prompt or model
- Compare two model outputs side-by-side on the same prompt
- Drafts saved per conversation — survive refreshes and tab swaps
- Markdown, math, code with syntax highlighting, Mermaid diagrams
- Resume the last conversation on every device automatically
- Cmd+K everywhere, Cmd+Shift+M to switch model, Cmd+Shift+K for Canvas
Canvas
A live preview pane for everything models output.
When a response contains HTML, SVG, a Mermaid diagram, or rich Markdown, Canvas opens a resizable preview to the right. Iterate on a landing page or system diagram without ever leaving the thread.
- HTML iframe preview with isolated sandbox
- SVG and Mermaid render natively, no third-party redirects
- Code view with one-click copy, preview view with one-click open in new tab
- Resizable divider, remembered between sessions
Tools
Web search and image generation that any model can call.
Tools are model-agnostic. Ask Claude to search the web, ask GPT to generate an image, ask Gemini to do both in the same turn. The model picks the tool; we route it.
- Tavily-backed web search with cited URLs in every answer
- FAL image generation and image-to-image editing
- Code interpreter (sandboxed Python) for charts and data work
- File uploads: PDFs, images, CSVs, source code — parsed into context
Memory
It remembers — and you control what it remembers.
Tell it once that you ship in Go, deploy on Fly.io, and prefer terse code review feedback. Every future conversation starts with that context. Inspect the full memory list at any time and prune what you don't need.
- Persistent across conversations and devices
- Inspectable list — see exactly what the model knows about you
- One-click delete on any memory
- Per-conversation overrides when you want a clean slate
Privacy
Your conversations are yours.
We don't train on your data, and we configure every provider integration with no-training, no-retention flags wherever they're available. Conversations are encrypted at rest and deletable in one click.
- Zero-retention provider configs where supported (OpenAI, Anthropic, Mistral)
- Encrypted at rest, TLS in transit
- Per-conversation and full-account delete
- Export everything to JSON whenever you want
Stop juggling tabs.
One subscription, every model. Start on the free tier and upgrade when you outgrow it.