Introduction
Project Name: ReBit – Decentralized Short Video Processing Engine
Name of the organization submitting the proposal: Dapp Mentors
Dapp Mentors is a two-time Sia Foundation grant recipient specializing in developer tooling and decentralized infrastructure. Our previous grants delivered SiaLearn and SiaPeopleLearn, both built on Renterd. This proposal is a deliberate, focused step forward. Following the Committee’s direction and the Foundation’s current funding focus, ReBit will be built directly on the indexd + Sia Storage SDK pathway, delivering something practical, free and useful for everyday content creators.
Describe your project
ReBit is an open-source, self-hostable video processing engine that transforms long-form videos into multiple short-form vertical clips. It is designed to be fully user-owned, privacy-preserving, and operable without mandatory platform fees or centralized cloud dependencies.
A user uploads a single long-form video; a podcast, interview, lecture, webinar, or recorded event. ReBit automatically analyzes the video and extracts the most engaging moments to generate 3–10 short-form clips optimized for TikTok, Instagram Reels, and YouTube Shorts.
The system will work as follows:
- Accept a long-form video as input (MP4, MOV, WebM, MKV)
- Transcribe the audio into a timestamped transcript using Whisper
- Analyze the transcript with a local LLM (via Ollama) to detect high-value moments, hooks, insights, emotional peaks, and key statements
- Map selected moments back to exact video timestamps
- Cut and extract short clips using FFmpeg
- Convert each clip into vertical 9:16 mobile-friendly format
- Generate captions and subtitle overlays
- Store all outputs (source video, clips, transcripts, and metadata) on the Sia Network
All project data, generated assets, and clip metadata will be stored on the Sia Network for decentralized persistence, using the official Sia Storage SDK and indexd for object tracking and retrieval.
Application UI
The following mockups illustrate the intended user experience across all key screens:
Screen 1: Authentication
First-time setup requires only an Indexer URL. Returning users authenticate with App Key and password. Sessions are valid for 7 days.
Screen 2: Upload
Primary entry point for creating new jobs. Users drag and drop a long-form video (MP4, MOV, WebM, MKV, up to 5GB / 2hrs) to start the processing pipeline.
Screen 3: Processing Pipeline
Real-time job tracker displaying step-by-step pipeline progress: Upload → Transcription → Highlight Detection → Clip Extraction → Rendering → Saving to Sia storage.
Screen 4: Projects Dashboard
Main workspace listing all uploaded videos with processing status, clip counts, and quick access to active or completed jobs.
Screen 5: Clip Review
Core value delivery screen. Each generated clip displays a video preview, transcript excerpt, timestamp range, virality score, and hook strength rating. Users can Approve or Reject individual clips.
How does the projected outcome serve the Foundation’s mission of user-owned data? What problem does your project solve?
Content repurposing is one of the most time-intensive workflows for modern creators, developers, and educators. The dominant tools for this today are Opus Clip, Descript, and Vidyo.ai which are fully custodial. Every uploaded video sits on a company’s cloud servers in plaintext. The provider can analyze, train on, or lose the content at any time. Long-form recordings of private interviews, internal product demos, or unpublished educational lectures carry real privacy risk when processed on third-party platforms.
ReBit will solve this by making the entire pipeline self-hostable and user-owned. All final outputs will be stored on the Sia Network; not on AWS S3, not on a third-party CDN, and not on a shared cloud bucket. The user holds the keys. Recovery will require only the operator’s BIP-39 mnemonic, from which the Sia App Key is derived.
ReBit will integrate with Sia exclusively through the official Sia Storage SDK and indexd stack. It directly addresses the Foundation’s stated interest in tools for “sharing/viewing large single files (videos, data sets, high-res photos)” and aligns with the SDK + indexd grant theme.
Are you a resident of any jurisdiction on the restricted list? No.
Will your payment bank account be located in any jurisdiction on the restricted list? No.
Grant Specifics
Amount of money requested and justification with a reasonable breakdown of expenses
The total requested amount is $10,000 USD, paid across a three-month development term against the milestone schedule below. The full amount covers development labor.
| Category | Detail | Amount (USD) |
|---|---|---|
| Development labor | One developer, three months; backend pipeline, Sia integration, frontend UI, Docker Compose stack, integration tests | $10,000 |
| Total | $10,000 |
High-level architecture overview. What security best practices are you following?
ReBit will be built as a modular, self-hostable system deployed entirely via Docker Compose. It will be composed of seven layers:
1. Video Ingestion Layer Will handle long-form video uploads and metadata extraction through a Next.js frontend. Accepts MP4, MOV, WebM, and MKV. File format and size will be validated before processing begins.
2. Transcription Layer Will convert audio to a timestamped transcript using OpenAI’s Whisper model running locally via FastAPI. No audio data will leave the user’s infrastructure.
3. Highlight Detection Engine Will analyze the transcript with a local Ollama-powered LLM to identify the most engaging and high-signal segments. Segments will be ranked by relevance, hook strength, and potential impact (no proprietary cloud AI API will be called).
4. Video Processing Engine An Inngest event-driven pipeline will map selected transcript segments back to video timestamps and use FFmpeg to extract precise clips from the source video.
5. Short-Form Rendering Layer Each clip will be converted to 9:16 vertical format, cropped, reframed, and captioned using FFmpeg. Output will be standard MP4.
6. Storage Layer (Sia Integration) All outputs (source video, clips, transcripts, and clip metadata) will be uploaded and pinned to the Sia Network using the official Sia Storage SDK. Each object will be registered with indexd for health monitoring and retrieval. No output will touch a centralized cloud bucket.
7. Indexing Layer Sia indexd will be used for metadata indexing, enabling search and retrieval of generated clips and projects.
Security practices:
- No user content will be sent to external APIs; all AI inference will run locally via Ollama and Whisper
- The Sia App Key will be derived from the operator’s BIP-39 recovery phrase and will NEVER be stored in the database or exposed in logs
- Application API keys will be stored as SHA-256 hashes and shown only once at creation
- Session tokens will expire after 7 days, requiring re-authentication
Technology stack:
| Layer | Technology |
|---|---|
| Frontend | Next.js |
| Workflow orchestration | Inngest (event-driven pipeline) |
| Backend / ML inference | FastAPI (Python) |
| Local LLM | Ollama |
| Transcription | Whisper |
| Video processing | FFmpeg |
| Storage | Sia Storage SDK + indexd |
| Database | PostgreSQL |
| Infrastructure | Docker Compose |
Goals and timeline for completion
The goal is a working open-source release of ReBit that an operator can clone, configure, and run locally to process long-form videos for absolutely free and store all outputs on Sia. Development is split into three monthly milestones.
Milestone 1: Foundational Stack, Ingestion, Transcription, and Highlight Detection (Month 1)
Objective: Establish the Docker Compose stack and prove the first half of the processing pipeline; from video upload through highlight detection.
Deliverables:
docker-compose.ymlbringing up PostgreSQL, Inngest, FastAPI, Ollama, and indexd under healthcheck-gateddepends_onchains- Next.js upload screen: drag-and-drop video upload with format and size validation, pipeline trigger on successful upload
- FastAPI transcription service: Whisper integration producing timestamped VTT/JSON transcripts
- Highlight detection engine: Ollama LLM prompt chain analyzing transcripts and ranking segments by hook strength, virality score, and emotional signal
- Inngest event pipeline wiring:
upload → transcribe → analyzestages with step-level status tracking persisted to PostgreSQL - Job status screen: real-time pipeline tracker displaying step-by-step progress (Upload, Transcription, Highlight Detection)
- PostgreSQL schema: jobs, transcripts, segments, and Sia object references
Acceptance criteria: A scripted run successfully uploads a long-form video, the transcription service produces a timestamped transcript, and the highlight detection engine returns a ranked list of candidate segments with scores visible in the job status screen.
Milestone 2: Clip Extraction, Vertical Rendering, and Sia Storage (Month 2)
Objective: Complete the video processing pipeline (from segment extraction through vertical rendering) and prove the Sia write path for all outputs.
Deliverables:
- FFmpeg clip extraction service: maps ranked transcript segments to video timestamps and cuts precise clips from the source video
- Short-form rendering: converts each clip to
9:16vertical format with automatic center-crop reframing and caption/subtitle overlay - Inngest pipeline continuation:
analyze → cut → render → storestages with step-level status tracking - Sia write path: each generated clip and the source video are uploaded and pinned to the Sia Network via the official Sia Storage SDK; all Sia Object IDs persisted to PostgreSQL
- indexd integration: each uploaded object registered with indexd for health monitoring and retrieval
- Job status screen updated to reflect the full pipeline: Upload → Transcription → Highlight Detection → Clip Extraction → Rendering → Saving to Sia
Acceptance criteria: After a full pipeline run, all generated clips and the source video appear as pinned objects on Sia, all Sia Object IDs are correctly mapped to their parent job in PostgreSQL, and the job status screen reflects completed status for all pipeline steps.
Milestone 3: Full UI, Auth, Export, Documentation, and Release (Month 3)
Objective: Deliver the complete user-facing application (auth, project dashboard, clip review, and export) along with documentation and an end-to-end integration test.
Deliverables:
- Auth screen: first-time indexer URL configuration and returning-user App Key + password login with 7-day session management
- Project dashboard: lists all uploaded videos with processing status, clip counts, and quick access to active or completed jobs
- Clip review screen: displays generated clips with video preview, transcript excerpt, timestamp range, virality score, and Approve/Reject actions
- Export: approved clips downloadable as standard MP4
- README with Docker Compose quickstart, environment configuration guide, and Sia indexd connection instructions
- End-to-end integration test: uploads a source video, runs the full pipeline, verifies all clips and source assets appear as pinned objects on Sia, and confirms clip metadata maps correctly to Sia Object IDs
Acceptance criteria: An external tester clones the repository, runs docker compose up, authenticates, uploads a long-form video, and receives multiple short-form vertical clips stored on Sia; reviewable, approvable, and exportable through the UI.
Who is the target user for your project?
The primary user will be an independent content creator, educator, developer, or anyone who produces long-form video content (podcasts, interviews, lectures, or recorded events) and wants to repurpose it into short-form clips fast, and without uploading sensitive content to a third-party platform or struggle to manually edit them on a video editor.
The secondary user is an operator deploying ReBit on their own infrastructure for a team, studio, or organization.
Both users will benefit from full data ownership: their source videos, transcripts, and generated clips will be stored exclusively on the Sia Network under their own cryptographic control, with no centralized platform holding plaintext content.
Plans for this project following the grant
The repository will be released under the MIT license and remain openly developed.
Following this grant, the natural next steps include:
- Multi-language transcription and clipping support
- Auto-publishing integrations for TikTok, Instagram Reels, and YouTube Shorts
- Collaborative clip editing workflows
- A hosted deployment for users who prefer not to self-host, the subject of a future Standard Grant proposal
Potential risks that will affect the outcome of the project
Whisper transcription accuracy and speed. For heavily accented speech or poor audio quality, transcript accuracy may be lower, which would affect the quality of highlight detection downstream. Mitigation: the UI will expose transcript editing so users can correct segments and regenerate clips.
Ollama LLM output consistency. Local LLM output for highlight detection may be inconsistent across different hardware configurations or model versions. Mitigation: structured JSON prompts with explicit scoring rubrics and a fixed temperature setting will be used; a keyword-based segment scoring fallback will activate if LLM output fails schema validation.
FFmpeg vertical reframing. Automatic 9:16 reframing for arbitrary footage may misframe some clips depending on subject positioning. Mitigation: center-crop reframing will be used as a reliable baseline default, and this limitation will be clearly documented.
Development Information
Will all of your project’s code be open-source?
Yes. The entire codebase will be released under the MIT License.
Repository: https://github.com/Dapp-Mentors/rebit (All codes will be pushed here)
Do you agree to submit monthly progress reports?
Yes. We will submit a monthly progress report here on forum.sia.tech covering each milestone’s deliverables.
Contact Info
Organization: Dapp Mentors
Email: [email protected]
LinkedIn: https://www.linkedin.com/in/darlington-gospel