Skip to content

Chat and message rendering

The conversation view normalizes every agent's output into one consistent UI while preserving the structured activity behind it.

Whichever agent is behind a session, AgentDock renders its output the same way: a conversation view built on GitHub-flavored Markdown, with syntax-highlighted code blocks and diffs, tables, and sanitized HTML — never raw, unsanitized output.

What's normalized

  • Markdown rendering — headings, lists, tables, and fenced code blocks render the same regardless of which agent produced them.
  • Structured activity — tool calls, file edits, and permission requests appear as readable cards interleaved with the assistant's text, rather than buried in log output. This is built on the shared agent event model that every adapter maps its own transport into.
  • Sanitization — rendered HTML is sanitized before display. Don't rely on an agent being able to execute arbitrary embedded HTML through the conversation view.

Why this matters across agents

Claude Code and Codex report structured, typed events from their SDKs; Antigravity has no SDK and is driven over a PTY session with its raw terminal output classified into the same event shapes. The result is that a session with Antigravity looks and feels like a session with Claude Code or Codex in the conversation view, even though the underlying transport is completely different — see Agent adapter architecture.

Related