Skip to content

Agents

Configure the executable path, permission mode, model, and reasoning effort for each connected coding agent.

The Agents tab in Settings is where AgentDock's per-agent configuration lives. Every supported agent — Claude Code, Codex, and Antigravity — gets its own row with the same four fields.

Fields per agent

FieldPurpose
customPathOverride the executable AgentDock launches for this agent, instead of relying on PATH detection. Useful when a CLI isn't on PATH, or you want to pin a specific install.
permissionModeThe agent's permission/sandbox mode — see Permissions for the exact modes each agent supports.
modelThe model the agent CLI should use, where the underlying CLI/SDK exposes a model catalogue.
reasoningEffortA per-agent effort/quality dial, where the underlying CLI supports it — see Models & reasoning effort.

Testing detection

Each agent row has a live Test action that re-runs AgentDock's detection logic against the configured path (custom or PATH-resolved) and reports back whether it found a runnable executable, its resolved path, and its version. If a CLI isn't found, see Agent CLI not detected.

Note

AgentDock does not install or manage authentication for any agent CLI. You install and sign in to each one yourself — see Authentication.

Diagnostics

The Agents tab includes an Open Diagnostics link that jumps to the Advanced tab, where AgentDock surfaces detailed detection and environment information (app/Electron/Chrome/Node versions, database path, and Codex runtime resolution details) — useful when collecting diagnostic information for a bug report.