Camera workspace
A place for the whole setup.
Organise cameras by network and local groups, inspect health and manage supported arm/disarm and motion controls. Device commands carry pending states and observed read-back rather than assuming success.
← Back to workAI & automation / Local-first software
A little more light on what happened.
Porchlight
Your cameras.
Your computer.
Your context.
A free, open-source companion for Blink cameras. Pair compatible Sync Module local storage with a PC archive and local AI, without a monthly Porchlight subscription or paid AI API.
Open source · Windows development build
No monthly app subscriptionSync Module local storageAI on your own PC
More from the hardware you own
Keep the clips.
Skip the monthly plan.
With compatible cameras and working Sync Module local storage, a paid Blink plan is optional. Porchlight can retrieve those recordings, sync them into your PC archive and add local AI analysis, searchable descriptions and owner-reviewed face recognition.
Recording sourceSync ModuleCompatible local storage
→Your storagePC archiveSync, play, search & keep
→Your processingLocal AIOllama + optional face recognition
- No Porchlight subscriptionFree, open-source software running on your hardware.
- No paid AI APIAnalyse archived video with a locally installed model.
- Keep your own archiveUse local disk space, retention controls and backups.
Requires compatible cameras, a supported Sync Module and working removable storage, such as USB on Sync Module 2 or microSD on Sync Module XR. Retrieval depends on hardware support and Blink connectivity; Sync Module Core has no removable local storage. Check Blink storage compatibility ↗
The product idea
A camera sees a moment.
You need the context.
Porchlight connects the everyday camera experience with the things that happen after a clip is saved: finding it again, understanding the visible activity, reviewing a possible match and deciding what to keep.
Recording archive
More than a stream of alerts.
Bring archived Blink clips, manual MP4/JPEG imports, live recordings and snapshots into one date-grouped Activity view. Play, export, favourite, annotate and correct what you keep.
Search & recall
Find the moment worth keeping.
Search notes and descriptions, filter observed categories and colours, save useful views or preview natural-language filters before applying them. On-demand daily digests collect the day’s saved context.
Everyday operations
Useful after the first demo.
Notification rules, quiet hours, snooze, a local inbox and optional browser push sit alongside retention previews, verified backup/restore and separate worker diagnostics.
From recording to insight
Preserve first.
Understand next.
The archive and the analysis have different jobs. Explore how a recording moves through local processing without making AI a prerequisite for keeping the original.
Illustrated workflow · fixed example
Illustration only · no camera connection
Archive
The original comes first.
On a compatible no-plan setup, clips are retrieved from Sync Module local storage into the PC archive. Blink cloud clips and manual imports can also enter the archive. Durable downloads keep preservation separate from AI analysis.
Sync Module / Other sources → Local media archive
A model failure does not need to become a lost recording.
Frames
A clip becomes a set of moments.
FFmpeg extracts bounded frame samples. Light and Balanced workloads trade resource use against coverage. Sample times and frame hashes make accepted evidence reviewable.
Local recording → Timestamped image samples
Sampling can miss brief activity. The original remains the reference.
Ollama
Useful observations, on your PC.
A local Ollama vision model returns structured observations. Schema and evidence checks validate the response before it becomes a saved analysis revision.
Selected frames → Validated local observation
No cloud inference fallback or paid AI key is required.
Review
The owner has the last word.
Open the sampled evidence, play the original, correct the description and add labels or notes. Face recognition remains a separate, optional source of possible matches.
Observation + Original → Owner-reviewed context
A description or possible face match is an aid to review, not proof.
Optional local face recognition
A familiar face.
A considered match.
Activity descriptions and identity suggestions are separate systems. Ollama describes what is visible; YuNet and SFace support an owner-reviewed recognition workflow.
Detect locally
Optional YuNet detection finds faces in sampled frames and surfaces timestamps, boxes and image-quality flags. Automatic detection starts off.
Name with intention
The owner reviews source recordings and assigns selected clear examples to a person. Suggested groups remain tentative until checked.
Suggest a familiar face
SFace compares later samples against locally saved exemplars. Uncertain or low-quality comparisons stay unnamed, and possible matches remain reviewable.
Possible match. Always reviewable. Names begin with the owner’s selected examples. Suggestions do not establish identity, and recognition does not automatically disarm a camera or trigger an emergency action.
One workspace, different screens
At the desk.
In your pocket.
A responsive interface keeps Activity, Faces, Live View and Settings within reach. Home-screen installation and optional browser push extend the local workspace to a phone.
- Search, filters and saved views in a compact toolbar
- Recorded clips, snapshots and clear source labels
- Rules, quiet hours, snooze and a notification inbox
- Optional authenticated remote access with explicit HTTPS setup
Phone access depends on the host PC being awake and reachable. Browser push uses browser-vendor delivery infrastructure.
The system underneath
A web app.
A local engine.
One Windows-native service brings the interface, jobs and archive together. No hosted database, Docker or cloud AI service is required for the core installation.
Blink servicesCloud-dependent camera access
On the owner’s PC
React / TypeScript / Vite↕Python / FastAPISame-origin API · Authenticated updates · Durable jobs
SQLite + MediaArchive & job state
FFmpegBounded frame samples
OllamaLocal vision analysis
OpenCVYuNet + SFace
Durable work, independent failures.
SQLite-backed jobs separate discovery, downloads, AI and face detection. Leases, retry limits and restart recovery let the archive continue when a model or optional worker is unavailable.
Evidence travels with the observation.
Frame selection records timestamps and hashes. Schema and evidence validation guard accepted observations, while sampled-image review connects a description back to the original.
A native Windows home.
FastAPI serves the Vite build from the same origin. SQLAlchemy and Alembic manage local SQLite data; FFmpeg handles bounded media work. Native Ollama stays on loopback.
Private data has a local lifecycle.
Authenticated playback, CSRF-protected changes and Windows DPAPI-backed secret storage sit alongside previewed retention and backup checks. SFace exemplars are encrypted for the Windows user.
Clear product boundaries
Local AI.
Connected cameras.
Porchlight is an independent Blink companion. The recording archive and AI processing are local, while new Blink retrieval and live sessions still depend on Blink’s services.
Live View, recording, snapshots, audio and camera controls depend on device capability. Live sessions are bounded, with no two-way talk or continuous recording claim. Subscription-free retrieval depends on eligible hardware and verified recording access.
AI descriptions and face suggestions can be wrong or miss brief activity. The original recording and owner review stay central to the experience.