Agents overview
AgentDock gives Claude Code, Codex, and Antigravity a single, consistent chat interface.
AgentDock is a unified chat interface for coding-agent CLIs you already have installed. It doesn't provide model access, accounts, or subscriptions of its own — it detects the CLIs on your system, spawns them, and renders their output as a proper conversation UI instead of a raw terminal.
Supported agents
| Agent | CLI package | Notes |
|---|---|---|
| Claude Code | claude | Backed by @anthropic-ai/claude-agent-sdk; live model and permission-mode switching. |
| OpenAI Codex | codex | Backed by @openai/codex-sdk; live model catalogue fetched from the CLI's app-server. |
| Google Antigravity | agy (also probes antigravity, google-antigravity) | Terminal/PTY-based integration with output classification; no structured JSON transport. |
AgentDock detects each CLI on your PATH (plus a few known install locations on Windows and macOS) at runtime. You install and authenticate each CLI yourself, following that provider's own instructions — see Authentication.
What's consistent across agents
Regardless of which agent is behind a session, AgentDock gives you the same conversation view: rich Markdown rendering, syntax-highlighted diffs, structured tool-call and permission activity, and image attachments where the underlying CLI supports them.
- Agent configuration — custom executable paths and detection diagnostics.
- Models — per-agent model and reasoning-effort selection.
- Permissions — each agent's real permission-mode identifiers.
Handoff between agents
Within a session, AgentDock can build a mechanical summary of the requests made, files touched, and open issues so far, then let you continue the conversation in a different agent. This is useful when one CLI stalls or you want a second opinion without re-explaining context by hand. See Session persistence for how this fits into a workspace's sessions.
You can also run the same prompt against multiple agents side by side — see Compare Mode.