Saved on 2026-04-19.
The core demo path is working:
- voice-first orchestration through OpenAI Realtime
- Codex running as the coding worker
- visible Codex progress and completion summaries
- JetBrains/WebStorm embedding
- IDE diff opening for Codex edits
At this point, the remaining work is polish on the Codex conversation panel, not product expansion.
Codex messages should render as real markdown in the conversation panel, including:
- headings (
#,##) - lists
- links
- bold
- inline code
- code fences
This should apply while messages are streaming, not only after the final message lands.
When Codex emits an edit/update message, show a small inline diff preview directly in the Codex chat.
Target behavior:
- green additions
- red deletions
- capped preview size
- roughly 5 added lines max
- roughly 5 removed lines max
- truncate anything beyond that
This should be sourced from structured file-change data, not scraped back out of prose summaries.
Ship the current system with:
- natural voice request
- Codex background execution
- visible progress
- IDE-native diff visibility
- polished Codex chat rendering
That is enough.