idapt launches: 200+ models, real files, and a whole team on one price

I built idapt because I wanted AI that works with my files, not a chatbot with an upload button.

Every AI app I used had the same shape: ask a question, get an answer, close the tab. The answer is gone. The files live somewhere else. The next conversation starts from zero. The models kept getting smarter and the workflow stayed a dead end. Single-vendor apps are good at what they do; the shape is the problem.

A workspace, not a chat window

idapt is where your models, files, agents, and computers work as one system:

• 200+ AI models: GPT, Claude, Gemini, Grok, DeepSeek, and the rest of the catalog. Switch mid-conversation, run one prompt across several models in parallel tabs, or let the auto-router pick.
• Drive: real files and folders that persist across every chat. The AI reads them, writes them, organizes them; nothing evaporates when the tab closes.
• Agents with a Knowledge Base: one per workspace. Folders, wikilinks, tags, and a graph your team curates, and what you teach it stays taught across conversations and models. It is the agent's persistent memory, not a prompt you re-paste.
• Cloud computers: full machines the AI operates from chat. Install packages, run servers, deploy code.
• idapt Code: the same agent loop pointed at your repositories, on any model you choose, with scheduled runs and cloud machines. Keep your terminal tool if you love it; idapt is also an MCP server that plugs into it.

Proof you can check

Most product pages ask for trust. Ours ask for a click.

The rankings are raw named benchmarks with provenance per lane, not an invented composite. Evals publishes real captured outputs with cost and latency run logs. Demos are real workspaces published read-only: the chats, the files, the artifacts, each with its run log attached.

When a claim here has a number behind it, the number is rendered from the same registries the product runs on, not from this post's copy. Where there is nothing real to show, the slot hides instead of showing a placeholder.

One price, the whole team

A subscription belongs to a Team, and one price covers its roster: Pro at $199/month covers five people; Max at $1,499/month covers fifteen. There is no trial: your first month comes back as credits ($100 on Pro, $750 on Max), so the plan pays part of your first month back while you judge the product on your own workload.

Before you assume a subscription anywhere is the right shape, do the arithmetic: the cost calculator prices your actual workload at list prices, and the gateway shows per-token cost on every request. If your volume is small, paying for usage can beat paying for a seat.

Get started

Two things worth doing in your first five minutes:

1. Import your ChatGPT history: one ZIP upload brings every conversation, branches included, into a folder next to your work.
2. Send one question to three models and compare the answers side by side.

Start free, and judge the difference in an afternoon.