ReelBrain v0.1.1 — Final Build Week submission

ReelBrain v0.1.1 — Final Build Week submission

ReelBrain v0.1.1 is the final OpenAI Build Week submission release of the source-available, local-first AI video-editing agent team.

Demo and submission

• Three-minute product walkthrough
• ReelBrain on Devpost
• Landing page

Final release highlights

• Four configurable editorial agents for story, retention, style, and continuity, coordinated by the Showrunner.
• Visible agent fan-out, orchestration rationale, tool activity, structured approvals, and render progress.
• Creator-controlled Like, Dislike, and Skip loops connected to inspectable Taste Profile evidence.
• Versioned, non-destructive local revisions with human approval before rendering or generated-tool deployment.
• ReelBrain Python runtime modules bundled inside the macOS app; creator state is stored under ~/.ReelBrain rather than a source checkout.
• Public demo, Build Week submission, rendered Ouroboros/Codex development flow, and commercial-license contact included in the repository.

Ouroboros and Codex with GPT-5.6

Ouroboros turned the initial product idea into a validated PM Seed through interview, clarification, execution, and evaluation. Codex with GPT-5.6 then implemented and debugged the Python runtime and Tauri desktop product, fanned out bounded engineering and editorial tasks, exercised the real interface, and kept product decisions aligned with the evidence and governance model.

Primary Codex development session: 019f7e65-0f11-7333-b3fe-c3d8401e0e2a

Codex owns ephemeral agent scheduling and synthesis. ReelBrain owns durable trust: source and taste snapshots, capability grants, evidence, approval, accepted plans, renders, and creator-controlled memory. Memory remains a behavioral prior, never source evidence.

Install on Apple Silicon macOS

• Installer: download ReelBrain-v0.1.1-macos-arm64.pkg and open it to place ReelBrain in Applications.
• Portable runner: download ReelBrain-v0.1.1-macos-arm64.zip, unzip it, and open ReelBrain.app.

The app bundle is ad-hoc signed with hardened runtime enabled, and strict bundle verification passed. The PKG installer is not Developer ID signed or Apple-notarized.

If macOS blocks the unsigned PKG

1. Click Done on the warning. Do not move the installer to Trash.
2. Open System Settings → Privacy & Security.
3. Scroll to Security and find the message saying ReelBrain was blocked.
4. Click Open Anyway.
5. Authenticate with Touch ID or the Mac password, then confirm Open.
6. Complete the normal macOS Installer steps.

Only use Open Anyway for a package downloaded from this official ReelBrain GitHub release. You can compare the package checksum below before continuing.

Host prerequisites remain Python 3.11+, FFmpeg, FFprobe, and either Codex on PATH or the Codex binary included with ChatGPT for macOS.

Portable app ZIP SHA-256: 985d4554837659218179f84e0db005102a24094cbabc775ffd0ad2b75daa9970

PKG installer SHA-256: cbca9e4f2d1e419e1220fd575bc98b36d36ff6281ca056c3104cb4c23c52b9e5

Verification

• Python: 284 tests passed.
• Rust: 14 tests passed; 2 live-account tests intentionally ignored.
• TypeScript and Vite production build passed.
• Packaged Python bridge smoke test passed from inside ReelBrain.app/Contents/Resources.
• codesign --verify --deep --strict passed for ReelBrain.app.

Current limitations

• Apple Silicon macOS is the validated release target.
• The installer is unsigned and the app is not notarized.
• Python, FFmpeg/FFprobe, and Codex or ChatGPT are not yet bundled.
• Caption accuracy still requires creator correction or an independent reference.
• Arbitrary timeline edits require additional governed tool capabilities.
• Direct social publishing is not included.

License

ReelBrain is source-available under the Sustainable Use License. The repository LICENSE file is authoritative. Commercial licensing inquiries may be sent to jqyu.lee@gmail.com.

ReelBrain v0.1.0 — Governed creator editing workspace

ReelBrain v0.1.0 — Governed creator editing workspace

ReelBrain v0.1.0 is the first source-available Community Edition release of the local-first AI video-editing agent team for creator-review drafts, explicit taste learning, and governed agent execution.

Demo

• Three-minute product walkthrough
• OpenAI Build Week submission on Devpost

Highlights

• Four editable editorial personas—Story, Retention, Style, and Continuity—coordinated by a Showrunner.
• Local project import, video playback, timestamp mentions, fullscreen review, non-destructive revisions, and version history.
• Rich creator chat with visible orchestration rationale, agent activity, tool calls, approval cards, and render progress.
• Structured Yes/No approval before revisions and human approval before generated tools can be built or deployed.
• Like, Dislike, and Skip review loops connected to an inspectable Taste Profile and evidence ledger.
• Grounded fan-out with frozen source catalogs, scoped taste snapshots, capability grants, workflow epochs, and accepted-plan validation.
• Korean and English caption artifacts in the preparation workflow.

Built through Ouroboros and Codex with GPT-5.6

ReelBrain started as an Ouroboros PM interview. The interview clarified the creator, editing outcome, acceptance criteria, memory philosophy, and governance boundaries, then generated a validated Seed. Codex with GPT-5.6 used that Seed to implement the Python runtime and Tauri desktop application, fan out bounded engineering questions, debug real media and UI failures, run verification, and preserve architectural continuity across the product loop.

Primary Codex development session: 019f7e65-0f11-7333-b3fe-c3d8401e0e2a

Ouroboros development fan-out used advisory lanes such as code-context inspection, contrarian review, simplification, architecture implications, and acceptance guarding. ReelBrain’s creator workflow uses four stable editorial personas. In both cases, independent agents return bounded proposals to a root orchestrator; they do not directly acquire durable authority.

Codex owns ephemeral scheduling, concurrency, retries, and synthesis. ReelBrain owns durable trust: the frozen source and taste snapshots, capability grants, evidence, validation, approvals, accepted plan, render version, and creator memory. Multiple tool trajectories may be valid, so outcome quality determines editing quality while sequence assertions are reserved for required safety and evidence steps.

Install and verify

For the easiest installation, download ReelBrain-v0.1.0-macos-arm64.dmg, open it, and drag ReelBrain into Applications. The DMG is the installer.

ReelBrain-v0.1.0-macos-arm64.zip contains the portable ReelBrain.app runner for users who prefer not to use a disk image. Unzip it, move the app where you want it, and open ReelBrain.

This initial build targets Apple Silicon, is ad-hoc signed, and is not notarized. macOS may require Control-clicking ReelBrain, choosing Open, and confirming the first launch.

The app bundle includes ReelBrain’s Python runtime modules and stores creator state in ~/.ReelBrain. It still requires Python 3.11+, FFmpeg, FFprobe, and either Codex on PATH or the Codex binary included with ChatGPT for macOS.

Portable app ZIP SHA-256: 92079e861ca1e7ee7fd4da6fca5e76c6b3a7d92ad300d07c6502b6cbcb6784de

DMG installer SHA-256: a1fa3885a5544cbc0c51a54ee8847a4d10a53e22f6c0680d485fa4ade214a155

Developers can run the source build with:

git clone https://github.com/Q00/ReelBrain.git
cd ReelBrain
uv sync --dev
uv run reelbrain doctor

cd desktop
npm install
npm run tauri dev

Known limitations

• macOS on Apple Silicon is the validated baseline; other platforms are not yet validated.
• The release is not Apple-notarized.
• Arbitrary timeline edits, caption rewrites, music replacement, or new reframing require additional governed tool capabilities.
• Caption accuracy still needs creator correction or an independent reference.
• Direct social publishing is not included.
• Managed rendering, synchronization, backups, updates, and creator support are planned Creator Cloud conveniences, not restrictions on the local Community Edition.

License

ReelBrain is source-available under the Sustainable Use License. The repository LICENSE file is authoritative.