Captain Code goes open source, using Jev to pick a model per prompt
The open-source terminal tool Captain uses Jev to classify tasks in milliseconds, then routes them to Claude Code, Codex, Grok, Cursor or open-weight models by quality, cost and latency, logging results locally.
Today we're open-sourcing #CaptainCode 🚀 🫶 Captain picks the right AI model for each prompt and runs it from one terminal, on your machine. It drives the @claudeai Code, @OpenAI Codex, @xai @grok and @Cursor CLIs you already pay for, and gives you access to the most capable open-weight models through @OpenRouter, @nvidia NIM and @huggingface. Why we built it? We no longer knew which model to use for what, we were under-using our subscriptions, and we wanted to hand more work to cheaper, capable open models. More than 25% of my prompts go to open-weight models as a result, and counting! How does Captain work? It leverages the newly released @typesafeai Jev model to triage every task in milliseconds. Then it routes on quality, cost and latency, with the outcome logged locally for iterative improvements. When a model hits its usage limit, it reads the reset time and carries the work, context included, to the next relevant available model. When you type /oss, the task goes to open models only. Learn more about Captain, try it now: https://captaincode.ai We welcome feedback and contributions 🖖 Crafted with ❤️ on @opencode.