Overview
Compare Mode runs the same prompt against multiple installed agents at once, in forced read-only sessions, so you can compare real responses side by side.
Compare Mode lets you send one prompt to two or three installed agents at the same time and see their real, live responses side by side — useful for judging which agent handles a given task best before you commit to one.
Read-only by design
Every pane in Compare Mode is forced into a safe, read-only permission mode automatically — Claude Code into plan, Codex into read-only, Antigravity into plan. This is deliberate: it stops multiple agents from writing to the same workspace at the same time. See Workspace behaviour for the full picture.
AgentDock Pro
Compare Mode itself is a free, core AgentDock feature — no Pro build required. AgentDock Pro adds one hardening on top: it auto-declines any prompt from an agent trying to escalate out of its forced read-only mode mid-comparison, and it automatically cleans up the temporary compare sessions when you leave Compare Mode instead of leaving them in your session list.