Three frontier models. One conversation that remembers.
Switch between GPT-5, Gemini and Claude mid-thread without losing context. Every conversation is stored, searchable and resumable — including the answer that was still streaming when you closed the tab.
No card to start · Sign in with Google, GitHub or LinkedIn · Delete everything in one click
- GPT-5 miniOpenAIThe balanced default. Reasons through multi-step work without the wait.Free
- GPT-5 nanoOpenAIThe fastest and cheapest. Best for short questions and quick edits.Free
- Gemini 3.1 ProGoogleLong-context reading. Strongest on large documents and mixed media.Pro
- Claude Sonnet 4AnthropicCareful, literate answers. The one to pick for code review and prose.Pro
- Streams resume after a reload
- Per-user data isolation on every query
- Social sign-in only — no password to leak
- Usage and spend visible to you, not just to us
Built like a tool, not like a chat box.
01 / threads
Conversations that survive the tab
A generation keeps running when your connection drops. Reload and the answer is still arriving — not restarted, not lost, the same stream reattached.
02 / memory
Memory you switch on yourself
Durable facts and preferences carried between conversations, stored per account and searchable by meaning. Off until you say otherwise, and erasable one fact at a time.
03 / models
Change model without changing thread
The picker swaps the model, not the conversation. Ask a cheap model first, escalate to a careful one when the answer matters, and keep every earlier turn in context.
04 / tools
Answers that go and look
Live product, weather, market and news data fetched mid-answer and rendered as real cards rather than a paragraph describing a table.
05 / attachments
Images and PDFs, read properly
Attach a document and the composer checks the selected model can actually read it before the upload starts — not after, with an error.
06 / ownership
Your data, exportable and deletable
Export any conversation as JSON or Markdown. Delete your account and every thread, message, attachment and memory goes with it, verifiably.
Pick the model per question, not per subscription.
Every model reads images and PDFs. When a provider goes down, the turn falls back to a healthy one at the same tier or lower — never to a paid model you did not pay for.
Compare the plansOn every plan
- GPT-5 miniThe balanced default. Reasons through multi-step work without the wait.
- GPT-5 nanoThe fastest and cheapest. Best for short questions and quick edits.
Added with Pro
- Gemini 3.1 ProLong-context reading. Strongest on large documents and mixed media.
- Claude Sonnet 4Careful, literate answers. The one to pick for code review and prose.
The questions worth asking.
- Do I need a credit card to try it?
- No. The free plan needs only a Google, GitHub or LinkedIn account, and it is a monthly allowance rather than a trial that expires.
- Which models can I use?
- GPT-5 mini and GPT-5 nano on the free plan; Gemini 3.1 Pro and Claude Sonnet 4 are added on Pro. You can switch between them inside a single conversation.
- What does the assistant remember about me?
- Nothing until you turn long-term memory on. Once enabled it stores short durable facts you can read, search and delete individually from the Memory Center.
- Can I get my conversations out?
- Yes. Every thread exports as JSON or Markdown from the sidebar, and deleting your account removes every thread, message, attachment and memory.
Open a thread and see.
The free plan is a month of real work, not a demo. Nothing to cancel if you stop.
Start free