Replace [DEEP RESEARCH] content tag with explicit mode and research_iteration fields

[GdoongMathew]

Summary

Deep Research mode and its iteration count were previously inferred from message content — a [DEEP RESEARCH] string was prepended to the message, and the backend counted messages to derive the iteration. This replaces both with explicit, typed request fields, removing all string-parsing and message-counting heuristics.

Backend

• ChatMessage gains mode: Literal["normal", "deep_research"] (default to "normal")
• ChatCompletionRequest gains research_iteration: int (defaults to 1, ge=1).
• simple_chat.py / websocket_wiki.py:
  • The active mode is now read directly from the last message (last_message.mode), which is guaranteed to be the user's current turn — no more scanning/stripping the [DEEP RESEARCH] tag.
  • The deep research iteration is now taken from request.research_iteration instead of counting assistant turns, so a brand-new deep research question always restarts at iteration 1 regardless of prior history.

Frontend

• ChatMessage / Message interfaces gain an optional mode field
• ChatCompletionRequest gains an optional research_iteration field.
• Ask.tsx now sets mode: deep_research instead of prefixing content with the tag, and sends research_iteration explicitly (1 on initial submit, the incremented value on auto-continue).

[gemini-code-assist]

Code Review

This pull request replaces the string-based [DEEP RESEARCH] tag detection with a structured mode field ("normal" or "deep_research") on chat messages across both backend and frontend files. However, there is a bug in the reverse traversal logic used to count the deep research streak in api/simple_chat.py and api/websocket_wiki.py. If a conversation contains a previous turn in normal mode, the loop will incorrectly increment the streak count before reaching the user message that should break the loop. Suggestions have been provided to traverse the messages chronologically to correctly track and reset the streak.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

There is a bug in the reverse traversal logic when counting the deep research streak. If the conversation contains a previous turn in 'normal' mode (e.g., User (normal) -> Assistant -> User (deep_research)), the loop will encounter the previous assistant message and increment current_streak before it reaches the previous user message to break the loop. This incorrectly inflates the iteration count.

To fix this, we can traverse the messages chronologically (forward) and track the active mode, resetting the streak if we exit deep research.

Suggested change

	current_streak = 0
	for msg in reversed(request.messages):
	if msg.role == "user" and msg.mode != "deep_research":
	break
	if msg.role == "assistant":
	current_streak += 1
	current_streak = 0
	in_deep_research = False
	for msg in request.messages:
	if msg.role == "user":
	in_deep_research = msg.mode == "deep_research"
	if not in_deep_research:
	current_streak = 0
	elif msg.role == "assistant" and in_deep_research:
	current_streak += 1

[GdoongMathew]

This should be fixed in 3e41900

[gemini-code-assist]

There is a bug in the reverse traversal logic when counting the deep research streak. If the conversation contains a previous turn in 'normal' mode (e.g., User (normal) -> Assistant -> User (deep_research)), the loop will encounter the previous assistant message and increment current_streak before it reaches the previous user message to break the loop. This incorrectly inflates the iteration count.

To fix this, we can traverse the messages chronologically (forward) and track the active mode, resetting the streak if we exit deep research.

Suggested change

	current_streak = 0
	for msg in reversed(request.messages):
	if msg.role == "user" and msg.mode != "deep_research":
	break
	if msg.role == "assistant":
	current_streak += 1
	current_streak = 0
	in_deep_research = False
	for msg in request.messages:
	if msg.role == "user":
	in_deep_research = msg.mode == "deep_research"
	if not in_deep_research:
	current_streak = 0
	elif msg.role == "assistant" and in_deep_research:
	current_streak += 1

[GdoongMathew]

This should be fixed in 3e41900