Cinestar - Privately Index and Search Through Media

As someone who's always drowning in screenshots, videos, and random media files, I built Clipwise to solve my own problem: finding that one specific image or video moment when I actually need it.

Initial version is targeted towards:

• Professional photographers/video editors who are slightly tech savy and comfortable using some ai inference tools

• Tech Savy people, who have accumulated lot of photos/videos/audios over time and need a better search tool
  

Demo video:

Cinestar features:

🔒 100% local - Your media never leaves your machine
🧠 AI-powered - Search by describing what you see ("red car in parking lot")
⚡ Instant results - No cloud delays, everything runs on your Mac
🎯 Precise timestamps - Jump to exact moments in videos

Some of it is a work in progress, let me know the issues

# Clipwise v0.1.3

TL;DR: Search your videos, images, and audio using natural language - now 40% faster!

## 🎯 What's New

### ⚡ Performance Boost

- 40% faster search during video processing (10s → 6s)

- Smart Ollama instance routing: dedicated resources for search, load-balanced for indexing

- No more waiting - search stays responsive even during heavy video processing

### 🎬 Video Intelligence

- Videos searchable in 60 seconds after upload

- Multi-modal understanding: combines audio transcription + visual keyframes + scene context

- Integrated video player with smart segment navigation

- Search by what you see AND what you hear: "person explaining code" or "sunset beach scene"

### 🐛 Fixes

- Native dependencies now work in packaged builds (Sharp, FFmpeg)

- Logo displays correctly in splash screen

- Smoother installation experience across all platforms

## 🏗️ Architecture

- 2 Ollama instances with nginx load balancing

- Batch-concurrent processing pipeline

- SQLite + vector search for blazing fast queries

## 🎬 UI Fixes

UI BUG showing alert dialogues. Sorry FOLKS!!

---

Try it: Download, run `docker-compose up -d`, and start searching your media library with natural language!

# Clipwise v0.1.46

TL;DR: Stable Production Version

# Cinestar (name change)

Release Date: October 11, 2025

## 🎯 Overview

## ✨ Major Changes

### 1. Simplifying Installation

• Removed dependency from docker, whisper is downloaded on request

• Transcription is no longer availalble via api, but a cli command

### 2. Speed up local folder indexing

• Local Folder connect indexing and searchability improved

• Fixes in search and captioning. Allowing for tagging as well

I understand, moving/downloading another new application, could be a hassle. Overall I would still want the option to use this privately hosted. The problem is with local, one needs to install any inference server, for example Ollama , is one of the easiest.

This seems to be a possible barrier to entry, not to mention, I can monetize it better. So would it be easier to deliver this via the web and a phone app?

That way say free tier users get only 5 videos or 30 minutes, have some data already setup for demo. On the app I can provide a toggle for `Incognito` mode, with Incognito on, it only relies on local inference servers, if not, BYOK or Use ours.

I really don't want to become a video/media provider, when I can rely on external integrations, so google, dropbox is something that is planned. So I don't have much motivation to store or ask users to upload their videos separately . Infact on a phone app things become much easier.

Any opinions?

Website url changed to: https://cinestar.sourceforge.io/

New Updates Coming for:

• Deduplication

• Organisation

• Option to use online models for people who can't run ollama and setup models

• Automatic download of models without manual intervention

• Internal Changes to handle data better.

This update focuses on smarter video understanding, faster performance, and a more flexible privacy model that gives you the choice between local and external AI providers.

🚀 Smarter Video Search

Video indexing is now multi-modal — combining audio transcription and visual scene understanding for precise, searchable context.

• Audio Phase (Whisper): 20-min videos transcribed in ~60s

• Visual Phase: Keyframe extraction + captioning for scene meaning

• Smart Segmentation: Every 30s segment gets its own searchable context

• Incremental Indexing: Start searching before processing finishes

It’s now easier than ever to find that exact moment in a long video — just type what was said or shown.

🖼️ Better Image Search

Images now go through multi-pass captioning — overview → details → context — giving the search engine richer understanding of each picture.

And we’ve opened up privacy settings:
You can now choose between 100% local processing via Ollama, or opt-in to external providers like OpenAI or Google Gemini for higher-quality captions and embeddings.

Drillbit stays local by default — but now you decide the balance between speed, quality, and privacy.

🎵 Audio Support

Audio files get full transcription and indexing too. Search podcasts, interviews, or voice memos by meaning — not filenames.
Waveform visualization and playback controls make navigation effortless.

🔍 Unified Search Experience

Search everything from one place — images, videos, and audio — with instant results.

• Hybrid Search: Combines AI semantic matching + keyword text search

• Context-Aware Results: Jump straight to relevant timestamps in videos or audio

• Deduplication: Smarter grouping of segments and parent files

• Real-Time Updates: Search results appear as you type

The experience feels like a private, offline version of “Google Photos + ChatGPT” — but built for your local drives.

🎨 Refreshed Interface (Driller v2)

A complete redesign focused on clarity and speed:

• Sleek, glassy theme with soft animations

• Scope filters (All, Folders, S3, Drive)

• Activity panel to track ongoing indexing

• Infinite scroll & smooth virtualization for large collections

Full-page media viewers replace old modals for a cleaner browsing flow — no z-index chaos, just focused exploration.

⚡ Performance Boosts You’ll Notice

• Sub-1s total search time

• ~100s for full AI indexing of a 20-min video

• 2–3s image caption generation

• 0.5s embedding generation

Everything feels faster and more responsive, even while indexing is running in the background.

🔒 Privacy — Now with Options

Previously, Drillbit was strictly offline.
Now, you can selectively enable external AI providers for tasks that benefit from cloud-grade models — while keeping everything else local.

Your data is never uploaded by default. External providers are opt-in, transparent, and reversible at any time.

🧭 Coming Next

• Cloud & NAS integrations (S3, Google Drive)

• Smart collections & auto-clustering

• Advanced filters (date, people, location)

• Batch tagging & export options

• PWA support

We’re focused on making Cinestar scale to 100K+ media files while keeping it private, fast, and personal.

Version: 0.1.64
Release: October 2024
License: MIT

Find anything — locally or with help from AI — while staying in control of your data.

Made with ❤️ for people who value privacy and want powerful local search