Skip to content

DarinShapiro/SentiHome

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

65 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

SentiHome

A vision-intelligence and rule-engine layer for Home Assistant. SentiHome adds VLM-based reasoning, conversational rule creation, multi-camera identity fusion, and feedback-driven optimization on top of your existing camera infrastructure — without locking you into a specific NVR.

Status: Pre-implementation. Architecture is documented and stable; code scaffolding in progress.


What it does

  • Watches your cameras through whatever NVR you already have (Agent DVR, Frigate, Blue Iris, Synology, QNAP, UniFi Protect) — or direct RTSP, no NVR needed
  • Reasons about what it sees using a local Vision Language Model with cloud fallback for hard cases
  • Creates rules conversationally ("Let me know when the mailman arrives and announce it on Sonos") — no YAML, no UI builders
  • Acts through Home Assistant — every notification, light, lock, speaker action goes through HA services
  • Learns from your feedback — automatically tests rule variants against archived clips when you dismiss alerts as false positives
  • Respects privacy — local-first by design, data classes enforced at the data-plane, GDPR/CCPA-aligned

Architecture at a glance

Cameras  →  NVR Adapter (pluggable)  →  Preprocessor  →  SentiHome Core  →  HA  →  Devices
                                                            ↓
                                                       VLM reasoning
                                                       Rule engine
                                                       Memory + learning

Full architecture: docs/architecture/ — 21 sections covering every component.

Start here: Architecture index


Repo layout

SentiHome/
├── docs/                    Architecture, requirements, decision logs
├── services/                Long-running Python services (core, VLM router, memory, etc.)
├── adapters/                Pluggable NVR adapters (one per platform)
├── ha-integration/          Home Assistant custom integration (Python)
├── frontend/                TypeScript: HA custom cards + optional operator dashboard
├── shared/                  JSON schemas, MCP protos, shared libraries (Python + TS)
├── infrastructure/          Docker compose, NATS config, DB migrations
├── tests/                   Integration + e2e tests
├── scripts/                 Setup, dev, release tooling
├── planning/                Source of truth for GitHub epics and issues
└── .github/                 Issue templates, CI workflows

Each top-level folder has its own README explaining scope, conventions, and pointers into the architecture docs.


Project status

  • Architecture: stable (see docs/architecture/)
  • Implementation: scaffolding phase
  • Target: v1 ships service-mode NVR adapters, direct-RTSP support, core VLM pipeline, HA integration, conversational rules
  • Long-term: v3–v4 makes the NVR layer optional (SentiHome absorbs motion detection, archival, clip generation)

Contributing

See CONTRIBUTING.md for development setup and conventions.

Open issues and epics live in GitHub Issues. Implementation work tracks against the epics defined in planning/.


License

See LICENSE.

About

Vision-intelligence and rule-engine layer for Home Assistant. Local-first VLM reasoning, conversational rule creation, multi-camera identity fusion, feedback-driven optimization.

Resources

License

Contributing

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors