Remove PythonInterpreter from Space prompt; note offline-only usage

Limit batched tool call inputs

Force max new tokens to 16384

Raise default max new tokens

Set default max turns to 50

Format tool observations like QUEST inference

Stop model before tool response boundary

Inject authoritative date into user prompt

Strengthen date-sensitive prompt guidance

Emphasize current date in system prompt

ui: refine example helper copy

ui: show full visited URLs

ui: fix Mind2Web 2 label

ui: remove moon landing example

ui: add Mind2Web2 RAG example

ui: clarify example prompt copy

ui: update max turns defaults and timeout copy

Update app.py

ui: remove walkthrough section per user request

walkthrough: scroll-driven replay (sticky state bar + active-step highlight)

ui: move walkthrough between Try examples and Output; rename label

ui: add "See it in action" walkthrough (real M2W2 trace, collapsed by default)

fix: keep question textbox intact when Stop cancels the run

ui: Paper card subtitle Blog -> arXiv (matches new arxiv URL)

fix: Stop button actually cancels run_ui generator (not the .then lambda)

ui: Paper card default URL -> arxiv abstract

css: drop both display headings one more size on mobile (1rem -> 0.9rem)

css: drop both display headings one size down

css: make the two display headings consistent (size + line-height)

css: shrink the two display headings on phones (<=600px)

css: one-per-row Open-release cards on mobile (fix char-wrapped labels)

css: stack action buttons full-width on mobile (fix squished Run Research)

css: fix mobile-stack selector (Gradio 5.x has no gradio-container / .gr-column)

css: stack the two-column layout on mobile/tablet (fix broken phone layout)

search: parse knowledgeGraph/peopleAlsoAsk; treat 0-result as ok+hint

perf: enable Gradio queue concurrency (fix "long queue" serialization)

ui: header title -> "QUEST: Training Frontier Deep Research Agents with Fully Synthetic Tasks"

ui: Paper card default URL -> the final QUEST github-page site

agent: bump default MEMORY_TOKEN_THRESHOLD 16k -> 80k

ui: drop "Visited Sources" appendix from final answer

ui: hero heading verb back to "research" (deep research framing)

ui: fix hero heading grammar; default Max Turns 6 -> 15

diag: surface underlying httpx/ssl cause + endpoint/model in extractor_error

agent: Azure OpenAI support for visit extractor + condenser

Delete blog_demo

agent: surface _llm_extract failure reason in visit tool_response

agent: wire research-repo Jina + LLM extractor + scholar + real condenser

ui: move connection-error tip from action-row to Output header

ui: auto-clear question after run; add connection-error retry tip

ui: capitalize QUEST in Space title, browser tab title, footer copy