Caspian picks the tool and model for each request, refines your prompt, and checks big changes before they run. On the plans you already pay for.
npm install -g @caspianmd/caspian
Free · Windows · macOS · Linux · Node 20.18+
Claude Code, Codex, Gemini, Cursor and Devin. Caspian runs them in the background and shows their answers in one session.
Quick questions go to the cheapest model, hard changes to the strongest.
Subscriptions before API keys. /use codex pins a tool.
Turns, tokens and time, next to the fast model. Dollars only when you're billed per token.
A cheap model fixes typos and turns "it" into what you meant. You choose what gets sent.
.caspian/context.md carries decisions and changes across tools and sessions.
Questions are read-only. Shell commands stay off until you press Shift+Tab.
The header shows how much of each tool's plan limits you've used.
Runs on your tools' own plans. Caspian Pro skips triage for clear, small requests.
Node.js 20.18+. The desktop app already includes it.
npm install -g @caspianmd/caspian
cd my-project && caspian
Claude Code, OpenAI Codex CLI, Gemini CLI, Cursor CLI and Devin CLI. caspian agents
shows which ones it found.
Nothing. Requests run on, and count against, the tools' own plans.
Among tools that are signed in, subscriptions come before API keys. Fast requests get the
cheapest model, capable ones the strongest, balanced ones stay on your current tool.
/use <tool> pins a tool; /use auto undoes it.
Questions never change files. Changes edit files directly and list what they touched. Big changes ask first.
Your request goes to the tool's own AI provider, as if you ran it yourself. With Caspian Pro, the request text also goes to Caspian's decision service.