4.1 KiB
Thinkloom Demo Script
Target length: 2 minutes 35 seconds. Keep the uploaded video under 3 minutes.
Recording requirements:
- Record the native Thinkloom window at 1920×1080 or 1440×900.
- Include clear spoken narration.
- Do not add music.
- Close notifications and hide personal information before recording.
- Do not show API keys, credentials, private paths, or the Devpost account.
- Avoid the Settings screen so the demo contains no unnecessary third-party branding.
Timed shot list and narration
0:00-0:15 — Problem and product
Screen: Open Thinkloom in Ideation with the sample project visible.
Narration:
Thinkloom is a native, local-first writing studio for people who want AI help without losing control of their ideas, prose, or creative history. It moves one writer through ideation, drafting, and finalization.
0:15-0:43 — Ideation and user control
Screen: Scroll the conversation and idea garden. Briefly show Accept, Reject, Variant, Edit, and drafting-set controls.
Narration:
In Ideation, conversation produces traceable idea suggestions. The writer explicitly accepts, rejects, edits, merges, archives, or creates variants. Source links preserve where an idea came from, and only chosen ideas enter the drafting set.
0:43-1:13 — Structured drafting and preview-first AI
Screen: Open Drafting. Select the accepted idea, show the structured manuscript, edit a sentence, and point to Draft a passage without sending a live provider request.
Narration:
Drafting keeps selected ideas beside a structured manuscript editor. Model output is preview-first: it is staged, can be partially accepted, regenerated, copied, or discarded, and never changes the manuscript until the writer chooses. Manual writing works with every model provider disabled.
1:13-1:38 — Finalization, history, and provenance
Screen: Open Finalization, then History and Provenance. Show the release checks, saved version, and contribution relationships.
Narration:
Finalization reuses the same preview-and-accept pattern for editorial changes. Named versions are restorable, and an append-only SHA-256 chain records meaningful creative events. The history explains relationships without claiming a simplistic human-versus-AI percentage.
1:38-1:56 — Export and privacy
Screen: Open Export and show Reading files, Project backup, and Authorship evidence.
Narration:
Thinkloom exports publication files, complete restorable backups, and sanitized evidence packages. Projects are local-first, cloud use requires explicit approval, credentials stay in the operating-system vault, and audio is never retained.
1:56-2:27 — Codex and GPT-5.6 collaboration
Screen: Keep the Thinkloom app visible. Optionally overlay a simple title card containing only text and dated commit hashes; do not show private chat content.
Narration:
I created Thinkloom during OpenAI Build Week with Codex using GPT-5.6 Sol at High reasoning effort. I set the audience, native-only boundary, privacy rules, provenance approach, and final product decisions. Codex turned my specifications into the interactive framework and implementation plan, implemented the React, Tauri, and Rust application, integrated structured editing and native persistence, fixed packaging issues, and ran the test and launch checks. When I rejected an early hosted companion, Codex removed it and consolidated the product into this native application.
2:27-2:35 — Close
Screen: Return to Drafting or the three-phase navigation.
Narration:
Thinkloom helps writers keep the thread—from first conversation to finished work—while every change remains theirs to review and decide.
Final review before upload
- Duration is less than 3:00.
- Narration is audible at normal volume.
- The app shown is the same build linked for judging.
- No music, private data, credentials, unrelated trademarks, or unlicensed material appears.
- The video is public on YouTube, not unlisted or private.
- The title and description identify Thinkloom and OpenAI Build Week without implying sponsorship or endorsement.