PLAN. EXECUTE.
REVIEW.
Dex decomposes your task, delegates execution to the coding agent of your choice, and reviews every change before it ships. Local models, cloud models, or your existing Claude Code / Codex subscription — mixed freely across every role.
Three roles, one pipeline,
zero babysitting
Dex decomposes a task with an orchestrator model, hands execution to a configurable coding agent — Claude Code, Codex, or a raw chat-API tool loop — and runs an independent review pass before anything ships. No single model babysits the whole job.
Every role can run on a different backend. Local Ollama for the executor, Anthropic for the orchestrator, Groq for the reviewer — mixed freely, checkpointed at every step, and resumable from any session.
One task. Three models. Zero blind spots.
Why trust one model's plan? In plan mode, the council fans your task out to multiple models at once.
Never reviewed by the model that wrote it. Self-review misses its own blind spots.
Orchestrate with a frontier model. Code on your own hardware. Review with another vendor.
Your terminal, but alive
Long agent run? Dex keeps you company while it works.
Rotating Hacker News feed line during execution. /vibes on and never stare at a spinner again.
/spotify play lofi · pause · next. Your soundtrack, one slash away.