Security teams and property managers have thousands of hours of footage but can only review it reactively after an incident, with no way to search by description
Cloud platform that ingests security camera feeds, uses still-frame detection to reduce costs on idle footage, and provides a search interface where users type descriptions like 'person carrying a package near back door at night'
Subscription tiered by camera count - $19/mo for 4 cameras, $49/mo for 16, enterprise pricing for larger deployments
This is a top-tier pain point. Every business with cameras has this problem: thousands of hours of footage that nobody watches until AFTER something goes wrong. Property managers literally scrub through hours of footage manually after incidents. The 425 upvotes and 107 comments confirm strong resonance. The pain is real, frequent, and currently unsolved for the SMB segment.
TAM is massive. ~70M+ security cameras in the US alone. Target segments: ~30M small businesses, ~300K property management companies, ~10K security firms, ~350K HOAs in the US. Even capturing 0.1% of SMB camera owners at $49/mo avg = $18M ARR. Global market multiplies this 3-5x. The camera-agnostic, cloud-native approach means no hardware TAM constraint.
Strong WTP signals: businesses already pay $20-$50/mo for camera cloud storage (Ring, Nest, Arlo). Adding search on top of storage they're already paying for is an easy upsell. Property managers and security companies have budget authority and clear ROI (faster incident resolution, liability reduction). The $19/mo for 4 cameras is a no-brainer price point. Knocked down from 8 because SMBs are notoriously price-sensitive and churn-prone, and you'll face 'I can just scrub the footage myself' objections.
This is where it gets hard. A solo dev can build a demo MVP in 4-8 weeks — ingest RTSP feeds, sample frames, run CLIP embeddings, store in a vector DB, build a search UI. BUT production-grade is much harder: (1) RTSP stream reliability across hundreds of camera brands is a nightmare, (2) still-frame/idle detection needs tuning per environment, (3) cloud video processing costs at scale are brutal without aggressive optimization, (4) real-time ingestion pipeline needs to be rock-solid 24/7, (5) data privacy/retention compliance (especially for multi-tenant SaaS with security footage). The still-frame detection idea is clever and essential for unit economics but adds complexity. A working demo ≠ a reliable product.
Clear gap exists: Spot AI and Coram AI offer NL search but target mid-market/enterprise at $100-200/camera/year with sales-driven onboarding. Camio is the closest but has stalled. Nobody owns self-serve semantic search for SMBs at $19-49/mo. The still-frame optimization is a genuine differentiator for unit economics. However, this gap exists partly because SMB security is a graveyard of startups — the segment is hard to sell to, hard to support, and has high churn. Spot AI could easily launch a self-serve tier and crush you.
Near-perfect subscription fit. Cameras run 24/7/365, footage is continuously generated, search value compounds over time (more history = more searchable). Natural per-camera pricing scales with customer growth. Retention should be strong once integrated — switching costs include re-ingesting historical footage and retraining staff. Storage and compute costs are ongoing, making subscription the only viable model. This is infrastructure, not a tool — once in, it stays.
- +Universal, visceral pain point — 425 upvotes confirms strong demand signal from a technical audience
- +Still-frame/idle detection is a genuinely clever insight that solves the #1 technical risk (cloud processing costs) and could be a defensible moat
- +Massive underserved SMB segment — incumbents are all chasing enterprise, leaving the $19-$49/mo tier completely open
- +Timing is perfect — vision-language models just crossed the accuracy/cost threshold to make this viable
- +Camera-agnostic approach means zero hardware risk and instant addressable market
- +Natural per-camera subscription pricing with strong retention characteristics
- !RTSP stream ingestion across thousands of camera brands/models is an operational hellscape — this is where most video SaaS startups die
- !Spot AI or Coram AI could launch a self-serve SMB tier in months and outspend you on acquisition
- !Cloud processing costs could destroy unit economics if still-frame detection isn't aggressive enough — you're one model accuracy regression away from burning cash
- !Security footage is extremely sensitive data — a single breach or compliance failure (GDPR, CCPA, BIPA) could be existential
- !SMB customers are expensive to acquire, hard to support (non-technical), and churn-prone — CAC payback periods in this segment are brutal
- !Camera connectivity issues will generate high support volume — 'why is my camera offline' will be 60% of your support tickets
AI-powered video intelligence platform with natural language search on security footage. Camera-agnostic, uses on-prem edge appliance for processing. Users can type queries like 'person in red jacket near entrance' to find clips.
Cloud-based physical security platform with AI-powered video search by attributes
Cloud-based semantic video search platform. Pioneer in making security footage searchable with natural language queries like 'delivery truck' or 'person with dog'. Works with existing cameras via lightweight gateway.
Video analytics platform known for 'Video Synopsis' — condensing hours of footage into short reviewable summaries. Strong attribute-based search, heat maps, crowd analytics. Owned by Canon.
AI-powered camera-agnostic video search platform. Natural language search over security footage. Focused on making existing camera systems intelligent without hardware replacement.
Week 1-2: Build a web app where users upload recorded footage files (MP4/AVI) — skip live RTSP entirely for MVP. Run still-frame detection to skip idle segments, extract keyframes, generate CLIP/SigLIP embeddings, store in Qdrant or Pinecone. Build a search UI with natural language input that returns timestamped clips. Week 3-4: Add basic RTSP ingestion for 2-3 common camera brands (Hikvision, Dahua, Reolink — covers ~60% of SMB cameras). Deploy on a single customer site. Week 5-6: Add multi-tenant auth, Stripe billing, camera management dashboard. Week 7-8: Onboard 5-10 beta customers from the HN thread commenters. Focus obsessively on search quality and RTSP reliability before scaling.
Free tier: upload up to 1 hour of footage for search (lead gen / demo). $19/mo: 4 cameras, 7-day searchable history. $49/mo: 16 cameras, 30-day history, email alerts. $149/mo: 64 cameras, 90-day history, API access, multiple locations. Enterprise: custom camera count, unlimited retention, SSO, audit logs, dedicated support. Upsell path: alert rules ('notify me when someone enters loading dock after 10pm'), analytics dashboards, incident report generation, integration with access control systems.
8-12 weeks to first paying customer if you start with file upload MVP and leverage the 107 HN commenters as a warm lead list. 4-6 months to $5K MRR if RTSP ingestion works reliably across common cameras. 12-18 months to $50K+ MRR requires cracking SMB acquisition (likely through security camera installer partnerships or property management software integrations).
- “Still-frame detection skips idle chunks, so security camera / sentry mode footage is much cheaper”
- “nobody can actually watch or review all of the video from those cameras”