SentrySearch

Free plan LinuxMacSelf-hostedWindows

What SentrySearch is

SentrySearch searches video footage by text or reference image and can return trimmed clips that match. It scans folders recursively for MP4 and MOV files, divides videos into overlapping chunks, and embeds the chunks alongside text or image queries in a shared vector space. Video vectors are stored in a local ChromaDB database. Search options include text, image, anomaly highlights, and optional reranking of candidate clips. Embedding choices include Gemini, Alibaba DashScope Qwen Cloud, LiteLLM, local Qwen3-VL, and MLX for Apple Silicon. The local backend runs on the user's machine without an API key and is described as private and offline-capable. With Qwen Cloud, local video chunks are uploaded to DashScope-managed temporary object storage for processing. Optional Tesla footage support can extract speed, GPS, and time metadata, and may add location labels through OpenStreetMap reverse geocoding. The project requires Python 3.11 or later and FFmpeg or imageio-ffmpeg. Local inference requires CUDA or Apple Metal, and macOS video decoding requires system FFmpeg. The repository identifies the license as Apache-2.0.

Who it suits

SentrySearch suits people who want to locate moments in local video collections using text, images, or anomaly highlights. Its local backend is an option for users who want processing on their own machine without an API key.

What it does well

Search supports text queries and reference images.
Scans MP4 and MOV files recursively.
Local backend runs without an API key.
Tesla footage can display metadata when available.

What to know first

Requires Python 3.11 or later and FFmpeg or imageio-ffmpeg.
Local inference requires CUDA or Apple Metal.
Still-frame detection can miss subtle motion.
Events spanning chunk boundaries may not match perfectly.

Limits

Still-frame detection is heuristic and may miss subtle motion; events that cross chunk boundaries may not match perfectly. The README estimates Gemini indexing at about $2.84 per hour of footage with default settings; local-backend calls use no API quota.

Getting started

Open the project's GitHub repository.
Install Python 3.11 or later.
Install FFmpeg or use the bundled imageio-ffmpeg package; on macOS, install system FFmpeg for video decoding.
Choose an embedding backend; local inference requires CUDA or Apple Metal.
Index MP4 or MOV footage from a folder.
Search with text or a reference image, or run highlights to rank anomalous clips.

Verdict

SentrySearch combines semantic video search with local indexing and several embedding choices. Review its processing requirements and the limitations around still frames and chunk boundaries before choosing a workflow.

Questions

Is SentrySearch free?

The project is free and open source under the Apache-2.0 license.

Which operating systems does it support?

The listed platforms are Linux, macOS, and Windows. It can also be self-hosted.

Can I search video without sending it to a cloud backend?

The local backend runs on the user's machine without an API key and is described as private and offline-capable.

What happens to footage with Qwen Cloud?

The official SDK uploads local video chunks to DashScope-managed temporary object storage before API processing.

What does SentrySearch require?

The README lists Python 3.11 or later and FFmpeg or imageio-ffmpeg. Local inference requires CUDA or Apple Metal, and macOS video decoding requires system FFmpeg.

Does it work only with Tesla Sentry Mode footage?

No. Its folder scanner finds MP4 and MOV files, including footage that was not recorded in Tesla Sentry Mode.

Facts from SentrySearch's own pages

Cloud data handling: For the Qwen Cloud backend, local video chunk files are uploaded to DashScope-managed temporary object storage before API processing. · github.com, 8 Oct 2026
Compatibility: The directory scanner recursively finds MP4 and MOV footage, including footage not recorded in Tesla Sentry Mode. · github.com, 8 Oct 2026
Embedding backends: It supports Gemini Embedding, Alibaba DashScope Qwen Cloud, LiteLLM gateways, and local Qwen3-VL models. · github.com, 8 Oct 2026
Highlights: The highlights command ranks anomalous clips in an index and can trim them automatically. · github.com, 8 Oct 2026
How it works: It embeds video chunks and text or image queries into a shared vector space, stores video vectors in a local ChromaDB database, and matches queries against them. · github.com, 7 Oct 2026
Integrations: SentrySearch can hand search results to the maker's SentryMerge tool and saved clips to SentryBlur. · github.com, 8 Oct 2026
License: The GitHub repository identifies the project as Apache-2.0 licensed. · github.com, 8 Oct 2026
Limits: Still-frame detection is heuristic, and search quality can be affected when an event crosses chunk boundaries. · github.com, 8 Oct 2026
Local processing: The local backend runs without an API key and processes footage on the user's machine. · github.com, 8 Oct 2026
Maker: The maintainer's GitHub profile names Soham Rajadhyaksha and lists Fremont, California. · github.com, 8 Oct 2026
Privacy: The README describes the local backend as private and says it runs entirely on the user's machine. · github.com, 8 Oct 2026
Privacy option: The local backend runs on the user's machine without an API key, which the README describes as private and offline-capable. · github.com, 7 Oct 2026
Purpose: SentrySearch performs semantic search over video footage and returns trimmed clips. · github.com, 8 Oct 2026
Requirements: The project requires Python 3.11 or later and FFmpeg, with bundled FFmpeg available through imageio-ffmpeg for the default setup. · github.com, 8 Oct 2026
Search methods: Users can search with text queries or reference images. · github.com, 8 Oct 2026
Search modes: It supports text search, image search, anomaly highlights, and optional reranking of candidate clips. · github.com, 7 Oct 2026
Supported footage: The directory scanner recursively finds MP4 and MOV files, including footage that is not from Tesla Sentry Mode. · github.com, 7 Oct 2026
Tesla overlay: An optional overlay can display Tesla dashcam speed, date, time, city, and road name when supported metadata is available. · github.com, 8 Oct 2026
Tesla support: An optional overlay extracts speed, GPS, and time metadata from supported Tesla driving footage and can add location labels through optional OpenStreetMap reverse geocoding. · github.com, 7 Oct 2026
Usage costs: The README estimates Gemini indexing at about $2.84 per hour of footage with its default settings and says local-backend calls use no API quota. · github.com, 7 Oct 2026
Video handling: It splits videos into overlapping chunks, stores embeddings in a local ChromaDB database, and can trim matching clips. · github.com, 8 Oct 2026

Where SentrySearch stands on Freedom251

AI Video Search Tools#9 of 24