feat: Render Markdown in chat and query output using rich

[KylinMountain]

Summary

• Use rich.live.Live + rich.markdown.Markdown to progressively render LLM responses with proper terminal formatting (headings, bold, code blocks, lists, etc.)
• Applied to both openkb chat (interactive REPL) and openkb query (single-shot)
• Falls back to plain text when --no-color is set (chat) or stdout is not a tty (query)
• Add rich>=13.0 as an explicit dependency

Test plan

• python -m pytest tests/ — 197 passed
• openkb chat — verify headings, bold, code blocks render with formatting
• openkb chat --no-color — verify plain text fallback
• openkb query "question" — verify formatted output
• openkb query "question" > file.txt — verify no ANSI codes in file

[KylinMountain]

[KylinMountain]

Code review

Found 1 issue:

1. In query.py, after live.stop() writes the tool call line to stdout, live.start() re-activates the auto-refresh thread (4 Hz by default). On the next refresh cycle, Live uses position_cursor() — which emits ANSI CURSOR_UP sequences based on the previous render height — to move back up into the already-printed area and overwrite it with the last Markdown renderable. This means the tool call text printed between live.stop() and live.start() will be overwritten/erased by the next Live refresh. The fix is to not re-start Live after a tool call (keep it stopped until more text arrives, then start a fresh Live), or to use live.console.print(tool_call_text) before stopping so it is incorporated into the live area.

OpenKB/openkb/agent/query.py

Lines 153 to 160 in 5d2b4b4

	args = getattr(raw, "arguments", "{}")
	if live:
	live.stop()
	sys.stdout.write(f"\n[tool call] {raw.name}({args})\n\n")
	sys.stdout.flush()
	if live:
	live.start()
	elif item.type == "tool_call_output_item":

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}

[KylinMountain]

Code review

No issues found. Checked for bugs and CLAUDE.md compliance.

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}

[rejojer]

Code review

Found 1 issue:

1. Duplicate text rendering after tool calls in both chat.py and query.py. The collected list accumulates all text deltas across the entire response, but Live instances are stopped and restarted around tool calls. When live.stop() is called, Rich commits the rendered Markdown to the terminal. A new Live is then created via _start_live(), but collected is never reset -- so the next live.update(Markdown("".join(collected))) re-renders all previously displayed text in the new Live region, causing visible duplication. This will happen any time the model produces text both before and after a tool call, which is a common agentic pattern.

chat.py -- collected is appended to at L277 and rendered at L280, but after the tool-call handler stops and restarts Live (L288-L302), the full collected content is re-rendered:

OpenKB/openkb/agent/chat.py

Lines 276 to 303 in 8cdc5a7

	need_blank_before_text = False
	collected.append(text)
	last_was_text = True
	if live:
	live.update(Markdown("".join(collected), code_theme="monokai"))
	else:
	sys.stdout.write(text)
	sys.stdout.flush()
	elif isinstance(event, RunItemStreamEvent):
	item = event.item
	if item.type == "tool_call_item":
	if last_was_text:
	if live:
	live.stop()
	live = None
	else:
	sys.stdout.write("\n")
	sys.stdout.flush()
	last_was_text = False
	raw = item.raw_item
	name = getattr(raw, "name", "?")
	args = getattr(raw, "arguments", "") or ""
	if live:
	live.stop()
	live = None
	_fmt(style, ("class:tool", _format_tool_line(name, args) + "\n"))
	live = _start_live()
	need_blank_before_text = True

query.py -- same pattern at L151-L167:

OpenKB/openkb/agent/query.py

Lines 150 to 168 in 8cdc5a7

	if text:
	collected.append(text)
	if live:
	live.update(Markdown("".join(collected), code_theme="monokai"))
	else:
	sys.stdout.write(text)
	sys.stdout.flush()
	elif isinstance(event, RunItemStreamEvent):
	item = event.item
	if item.type == "tool_call_item":
	raw = item.raw_item
	args = getattr(raw, "arguments", "{}")
	if live:
	live.stop()
	live = None
	sys.stdout.write(f"\n[tool call] {raw.name}({args})\n\n")
	sys.stdout.flush()
	live = _start_live()
	elif item.type == "tool_call_output_item":

Fix: use a separate display buffer that resets when Live is restarted, while keeping collected intact for the final answer assembly at L309.

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}