8.1 KiB
Date, Author, Tags
| Date | Author | Tags | ||||
|---|---|---|---|---|---|---|
| 2026-06-30 | Hermes |
|
Local Network Improvements
Self-hosted quick web preview service
Potential improvement: build a self-hosted Cloudflare Tunnel / ngrok-style service for quickly sharing local websites online through the family server.
Goal: from any local machine, publish a local web app or static preview to a clean HTTPS URL without needing an external tunnel account.
Proposed user experience
Run a command like:
publish-preview --name reyna-family --port 3000
Then get a public link such as:
https://reyna-family.preview.reynafamily.com
or a path-style link:
https://preview.reynafamily.com/reyna-family/
Recommended architecture
Use FamReynaServer as the public relay:
Local machine running preview site
→ outbound tunnel to FamReynaServer
→ Caddy HTTPS reverse proxy
→ public subdomain under preview.reynafamily.com
Recommended stack options:
-
FRP / frp
- Best for a reusable family preview platform.
- Supports multiple tunnels and subdomains.
- More full-featured than a raw SSH tunnel.
-
Rathole
- Lightweight, fast, simple reverse tunnel.
- Good if the goal is dependable simple tunnels with less config.
-
SSH reverse tunnel
- Fastest to implement because SSH already works.
- Good first prototype.
- Less polished for many named previews.
Safety features to include
- Random preview names by default.
- Optional password protection for previews.
- Auto-expire old previews after 7 or 30 days.
- Caddy-managed HTTPS.
- No direct exposure of local LAN services.
- Local machines only make outbound connections to FamReynaServer.
- Consider access logs and simple preview cleanup command.
Related context
FamReynaServer already runs Caddy and has existing reynafamily.com subdomains. Candidate preview domains discussed:
preview.reynafamily.comtmp.reynafamily.com
Initial recommendation: build with Caddy + FRP on FamReynaServer, with SSH reverse tunnel as the simplest prototype path.
Other discussed services not yet implemented
These are related infrastructure/service ideas that have come up in conversation and should stay on the local-network / family-server improvement backlog.
Podcast publishing service
Potential improvement: create a self-hosted podcast/audio publishing pipeline for family and ministry content.
Possible use cases:
- EMI teachings converted into audio-only episodes.
- Sermons or conference sessions published as podcast feeds.
- Kids nightly readings / family devotional mini-podcasts.
- Audio versions of selected YouTube or Subsplash recordings.
Possible service shape:
- Host RSS feeds and episode audio on FamReynaServer.
- Candidate subdomains:
podcast.reynafamily.com,podcast.emmint.com, oraudio.reynafamily.com. - Generate episode metadata from transcripts: title, speaker, date, topic, language, summary, scripture references, and links back to video.
- Optionally publish to Apple Podcasts / Spotify later, but keep the source of truth self-hosted.
Related future automation:
- Drop an MP3/WAV into an inbox folder.
- Automatically normalize audio volume.
- Generate transcript and summary.
- Produce show notes.
- Add episode to RSS feed.
- Optionally create a WhatsApp/shareable link.
Searchable media archive service
Potential improvement: build a searchable local archive for livestream recordings, teachings, transcripts, clips, and metadata.
Goal: make the video/audio archive reusable instead of only storing raw recordings.
Possible features:
- Store recordings on NAS / Jellyfin media storage.
- Generate transcripts for each teaching or event.
- Search by speaker, topic, scripture, language, date, event, or phrase.
- Link transcript segments back to video timestamps.
- Mark sections that are private or should not be published publicly.
- Support EMI content planning, YouTube publishing, podcast creation, and short clip extraction.
AI-assisted content repurposing service
Potential improvement: create a repeatable pipeline that turns long-form services or teachings into outreach assets.
Possible outputs:
- Full YouTube description and title.
- Shorts/Reels clip suggestions.
- Captions/subtitles.
- Quote graphics.
- WhatsApp-shareable short clips.
- Podcast/audio episode.
- Written summary or devotional article.
- Translation drafts.
This connects to the EMI media outreach strategy and the livestream/video generation process.
Open Design / local preview publishing bridge
Potential improvement: connect Open Design previews directly to the self-hosted preview service.
Goal: when Open Design creates a project, publish it automatically to a clean phone-friendly URL, for example:
https://reyna-family.preview.reynafamily.com
or:
https://preview.reynafamily.com/reyna-family/
This would avoid sending raw 127.0.0.1 or LAN-only preview links.
Kids voice gateway speaker identification service
Potential improvement: add local speaker identification before or inside the kids Hermes voice gateway.
Goal: identify whether the speaker is Grace, Elias, Adolfo, Nicole, or unknown, then tailor responses safely.
Recommended approach discussed:
- Run the model on a local server, not on the ESP32.
- Enroll short voice samples for each family member.
- Use speaker embeddings, likely SpeechBrain ECAPA-TDNN as the first practical model.
- Treat speaker ID as context only, not authentication.
- Unknown or low-confidence speakers should default to child-safe responses.
Kids ESP32 voice/image experience improvements
Potential improvement: continue improving the child-facing ESP32 assistant experience.
Possible features:
- Voice requests from kid devices routed to the separate
kidsHermes profile. - Short age-appropriate spoken replies.
- Image-aware interactions when a device sends a photo.
- Simple visual replies or animations on ESP32 screens.
- Parent-safe tool permissions and no admin/system tools exposed to children.
- Possible integration with local speaker identification.
Brain MCP / local knowledge service
Potential improvement: continue the PARA brain MCP server idea so local notes can be searched, read, and eventually written through a safer structured service.
Known open questions from the brain MCP project:
- Should it be a long-lived service or launched on demand by Hermes?
- Should write access start read-only/search-only first?
- How should Git commits and pushes be automated after writes?
- How can it preserve frontmatter, Obsidian links, and PARA filing rules?
Mac mini Voicebox voice generation service
Implemented in Reyna CLI on 2026-06-30 as reyna-cli voicebox and reyna-cli macmini voicebox.
Observed service state:
- Mac mini host:
192.168.68.102/Mac-mini-M4.local. - Voicebox app server process:
/Applications/Voicebox.app/Contents/MacOS/voicebox-server. - Voicebox REST API:
http://127.0.0.1:17493on the Mac mini. - Direct LAN access from the Pi to
http://192.168.68.102:17493was still refused during verification, so Reyna CLI tries the LAN URL first and falls back to SSH-local access on the Mac mini. - Verified commands: health, profiles, generate, generation lookup.
Useful commands:
reyna-cli voicebox health --json
reyna-cli voicebox profiles --json
reyna-cli voicebox generate "Hello from Reyna CLI" --profile Nicole --engine kokoro --json
reyna-cli voicebox generation <generation-id> --json
Potential future improvement: if Voicebox should truly be reachable directly from LAN, adjust the Voicebox server bind host or run a small localhost-only reverse proxy on the Mac mini that binds to the LAN interface with appropriate access controls.
Local service registry / subdomain map
Potential improvement: maintain a single local service registry for family infrastructure.
Goal: reduce scattered knowledge about which services run where.
Possible contents:
- Service name.
- Host machine.
- Internal IP and port.
- Public subdomain.
- Caddy reverse proxy target.
- Auth method.
- Backup/restore notes.
- Owner / intended audience.
- Whether it is private, family-only, EMI, or public.
This would pair well with Caddy, FamReynaServer, and the brain.