Agent
On a conventional operating system the shell takes commands. Here it takes intent. You describe the tool, the connection or the test you want, and it is written into your project as code you can read, edit, version and keep.
Bring your own engine
One engine contract, several engines behind it, switched per conversation. Claude Code and OpenAI Codex are the engines we have driven end to end, and only engines we have actually verified are offered as templates — a model you cannot vouch for is a dependency, not a feature.
It writes into your project
Output is files on disk, not a chat transcript you copy out. The agent edits in the same tree your editor and git panel are looking at, so every change is reviewable and revertible.
It knows what a backtest is
The research prompts carry the discipline: numbers that did not come from a real run must be labelled illustrative, fewer than about thirty trades proves nothing, and an unvalidated strategy is never described as tradeable.
It can call tools, not just write text
The agent drives MCP servers directly — running a backtest, rendering a surface, listing your algorithms — so it can do the work and then show you the result rather than describing what you should run.
- Do I have to pay for AI to use QuantIDE?
- No. Connect a CLI you already have and the agent runs on that. Hosted QuantIDE AI exists so you do not have to set anything up, not because the app is unusable without it.
- Is the code it writes mine?
- Yes. It is written to your disk in your repository. There is no runtime you have to stay subscribed to in order to keep running it.
- Can it build a connection to my broker?
- That is the normal way to use it. There is no fixed integration list — you describe the venue and the agent authors an adapter against its API as an MCP server in your project. See the Extensions layer.
Agent is part of QuantIDE. There is no separate purchase and no add-on — the download is free on macOS, Windows and Linux.
Download QuantIDE