Skip to content

PROJECTS.md

Latest commit

 

History

History
103 lines (88 loc) · 4.45 KB

PROJECTS.md

File metadata and controls

103 lines (88 loc) · 4.45 KB

Weekend Projects

Homelab projects to tackle when you have some free time.

Active Projects

lldap - Lightweight LDAP Server

  • URL: https://github.com/lldap/lldap
  • GitHub Issue: #162
  • Priority: Medium
  • Status: Not Started
  • Description: Deploy lightweight LDAP for centralized authentication across homelab services
  • Benefits:
    • Simple web UI for user management
    • Rust-based, low resource usage
    • Perfect for SSO across Proxmox, Grafana, etc.
  • Deployment Options: LXC container or Kubernetes

Proxmox CSI Plugin - Shared Storage for K3s

  • URL: https://github.com/sergelogvinov/proxmox-csi-plugin
  • GitHub Issue: #163
  • Priority: High
  • Status: Not Started
  • Description: Native Proxmox storage integration for Kubernetes PVCs
  • Context: Longhorn failed miserably, Crucible still promising but requires additional storage sleds
  • Benefits:
    • Direct integration with existing Proxmox storage (local-zfs, NFS)
    • No additional distributed storage layer overhead
    • Leverages infrastructure already in place

PageLM - Audio Podcasts from Blog Posts

  • URL: https://github.com/CaviraOSS/PageLM
  • GitHub Issue: #170
  • Priority: Low
  • Status: Not Started
  • Description: Generate audio podcasts from blog posts for "learning on the go"
  • Benefits:
    • Transforms existing documentation into audio content
    • Multiple TTS engine options (Edge TTS, ElevenLabs, Google)
    • Open-source NotebookLM alternative
  • Requirements: Node.js v21.18+, ffmpeg
  • Deployment Options: K8s, CI/CD integration, or on-demand

OpenMemory - Persistent AI Memory (Claude Integration)

  • URL: https://github.com/homeiac/OpenMemory (forked from CaviraOSS)
  • GitHub Issue: #171
  • Priority: High
  • Status: Implemented (MCP server running at localhost:8080)
  • Description: Give Claude Code persistent memory across sessions via MCP
  • Implementation:
    • Fork: homeiac/OpenMemory with LGM tools added
    • Tools: openmemory_lgm_store, openmemory_lgm_context + 5 base tools
    • Behavioral guidelines: CLAUDE.md "OpenMemory Integration" section
    • Test script: scripts/openmemory/verify-mcp-backend.sh
  • Behavioral Requirements:
    • R7: Auto-query context at session start
    • R8: Reinforce memories that help solve problems
    • R9: Fallback to CLAUDE.md when OpenMemory unavailable
    • R10: Ask before changing behavior based on memories
  • Storage Triggers: Issue resolution, discovery, user request, session end
  • Configuration: ~/.claude.json → mcpServers.openmemory
  • Usage: See CLAUDE.md "OpenMemory Integration" section for guidelines

OpenMemory Server - Cross-Project Shared Memory

  • URL: https://github.com/CaviraOSS/OpenMemory
  • GitHub Issue: #172
  • Priority: Medium
  • Status: Not Started
  • Description: Centralized OpenMemory server for multiple projects
  • Benefits:
    • Single instance serves home, chorus, devops repos
    • Web dashboard for memory visualization
    • Shared infrastructure knowledge base
  • Deployment Options: K8s with PVC or Docker on Proxmox LXC

OpenReason - AI Planning Validation

  • URL: https://github.com/CaviraOSS/OpenReason
  • GitHub Issue: #173
  • Priority: Medium
  • Status: Not Started
  • Description: Check and balance for Claude's planning and execution
  • Benefits:
    • 5-stage verification pipeline (classify → skeleton → solve → verify → finalize)
    • Catches missing steps, invalid reasoning, contradictions
    • Confidence scores before execution
    • Auto-repairs broken reasoning steps
  • Use Cases: Validate infra plans, check debugging hypotheses, verify refactoring

Backlog

Add future weekend project ideas here

Completed

Move completed projects here with completion date

Next Session - Frigate/Coral Decision

  1. Test face recognition: walk in front of cameras, check if Frigate recognizes faces (Asha/G) - if useful, keep on pumped-piglet
  2. If face recognition not useful: move Coral USB back to still-fawn, update deployment (node selector → still-fawn, hwaccel → vaapi), disable face_recognition
  3. Compare power usage: still-fawn (~25W total) vs pumped-piglet (~80-100W) with Frigate running