Run Claude Code, Codex, and Gemini side by side, each in its own context and folder. Fan out work in parallel, approve from your phone, and merge the best results into one.
A single agent in a terminal means one conversation, one folder, and you pinned to the desk to approve every command. Relay turns that into a control room: many agents, each sandboxed, all visible at a glance, and a decision is a tap away wherever you are.
Each context binds to a project folder and its own git worktree, so agents working in parallel never step on each other. Fan out four tasks, walk away, come back to four clean branches.
4 worktrees · 0 collisions
Pick two contexts, even ones running different engines, and merge them into a temporary review chat that carries the combined work. Compare approaches, take the best, throw the scratch away. The originals never change.
Claude Code, Codex, and Gemini, unified under the same streaming, the same approval flow, the same slash-command palette. Choose the right brain per task, switch it per context, never learn three tools.
Start a long task, pocket your phone, and get pinged the moment an agent needs a decision. Read exactly what it wants to run, tap Allow, and it keeps going. The desk is the cockpit; your phone is the pager.
Approve or deny every edit and command, with arrow-key option prompts, across all three engines.
Relay's commands plus each engine's own, in a fast arrow-key palette.
Drop a screenshot into any context and the agent sees it, on every engine.
A clean dashboard with a real dark theme and a distraction-free focus mode.
Signed, in-app updates for Relay and one-click updates for every engine.
No Relay servers. No telemetry. Your keys, your machine, your code.
Installs and opens with no security prompt. Self-contained, bundles its own runtime, and auto-updates after first launch. macOS (Apple Silicon).
curl -fsSL https://github.com/AkshobyaRaoSWE/relay/releases/latest/download/install.sh | bash