PROJECT_ENDGOALS.md

AI-Stalker — Project End Goals & What We’re Building

What this project is

AI-Stalker is an AI-powered, locally-run, fail-safe security monitoring platform that evolves AutoPTZ into a multi-node security orchestrator. The system is designed to coordinate security sensors (cameras, microphones, and other authorized devices) and produce real-time events and operator-visible context.

Non-negotiable principles (how contributors must frame the project)

• Consent-based, authorized use only. The platform is built for environments where the operator has permission to monitor and record (home/office security, personal assets, properly authorized camera placements).
• Fully local-first operation. Sensitive video and metadata should remain on the user’s hardware by default (offline-first). Cloud AI is optional.
• Zero paywall / no forced upgrade paths. No “pro” tiers for core functionality. No subscription gates for features, upgrades, or recovery.
• Reproducibility + auditability. Designs and integrations should be explainable, inspectable, and maintainable.
• Defensive engineering. We build safeguards and resilience—not “mischief.”

Contributors: when describing features, treat “repurposed / security appliance / failsafe node” as authorized, user-installed device automation for the app’s own local recording and monitoring—not anything involving unauthorized access.

Overall end goals

1) A real, dependable local NVR + security AI

• A true security workflow: ingest → analyze → alert → record → summarize → store.
• Works with USB cameras, NDI, and planned IP camera streaming (RTSP/ONVIF).
• Integrated PTZ automation (VISCA), coordinated with AI-driven tracking.

2) Advanced tracking and event understanding (all local)

• AI people tracking that stays stable across frames/cameras.
• Facial recognition + profiling workflows designed for operator control:
  • Trusted/ignored identities (to reduce noise and operator fatigue).
  • High-risk identities prioritized for attention.
  • Event-level summaries for what happened during a visit.
• Behavioral context tagging to provide operator awareness (e.g., distress, erratic movement, or falls) where the system can do so reliably.
• Alert categories for likely theft / violence / suspicious behavior patterns (implementation must be careful, conservative, and explainable).
• Context capture:
  • Who appeared (or stayed), and where.
  • When they arrived and when they left.
  • Who they interacted with (correlated identities).
  • What they brought / carried / removed (as supported by the implemented sensors + CV pipelines).

3) A multi-node “security hive” with automatic failover

• A leader/failsafe chain so recording and alerting continue if the primary node goes down.
• Automatic takeover with minimal or no user intervention.
• Configuration and relevant metadata/models are synced across nodes.
• Optional replication strategy (e.g., block-level or config+DB replication) to achieve resilience.

4) Idle-resource pooling across household computers

• The system should be able to use idle compute on other trusted household machines.
• It must never degrade normal usage:
  • Detect “idle” safely.
  • Activate only when idle thresholds are met.
  • Retreat immediately when the user resumes activity.

5) Full control for the operator (no paywalls)

• Core capabilities—failover, recording, AI inference, tracking, device management—must remain available without subscription.
• Updates and installer improvements must not require “pro” entitlements.

6) Security-first by design

• Encrypt sensitive data in transit and at rest where feasible.
• Add audit trails for configuration and admin actions.
• Keep integration boundaries clear and reduce supply-chain risk.
• Default retention should be minimal and configurable.

What we’re building (feature pillars)

1. Local AI NVR pipelines

   • Ingest: cameras/mics/sensors
   • Normalize streams
   • Run CV/AI inference locally
   • Record and index events

2. AI-assisted PTZ + tracking

   • Automated camera movement based on detections
   • Stable tracking logic and handoff behavior

3. Identity workflows

   • Trusted/ignored face handling to reduce false alarms
   • Priority tracking for high-risk identities

4. Event summarization

   • “Rundown while they visit” and “what happened when they left”
   • Interaction graphs (as supported by identity + tracking confidence)
   • Sensor-supported context (movement, crossings, objects carried where implemented)

5. Failover / continuity of service

   • Multi-node health monitoring
   • Promotion and takeover logic
   • Consistent recovery of configurations and state

6. Hive mode compute sharing

   • Idle scheduling and safe withdrawal
   • Distributed inference jobs to idle nodes

7. Authorized device expansion

   • Home automation integration (e.g., smart home workflows)
   • Repurposed, user-owned/installed devices as additional local sensors
   • Planned future: repurposed smartphones as auxiliary sensors (authorized use only)

“No wrong ideas” clarification (terminology guardrail)

When we use energetic or playful phrasing (e.g., “stalker,” “hive mind,” “security streaming device,” “repurposed e-waste”), the engineering meaning is:

• We are building a lawful, consent-based security system.
• We are automating only user-owned/authorized devices installed for this app’s monitoring.
• Failover means continued operation via additional user-installed nodes, not exploitation.

Target outcome

By the end state, AI-Stalker becomes:

• 100% free in all ways (no paywall features; no pro tiers).
• A local-first advanced security monitoring system.
• A resilient multi-node fail-safe NVR/orchestrator.
• An operator-friendly platform that turns video + sensor signals into useful, explainable security context—without unnecessary cloud dependency.