Brand-new features:
- The redesigned dark UI — full design-token theme (Space Grotesk/Manrope/JetBrains Mono, orange accent, panel-based
tables, console-style logs) replacing the stock Bootstrap look, plus new logo and branding
- EPG auto-mapping by channel name — upstream mapped only by tvg-id, so playlists without those tags (most IPTV
providers) mapped zero channels; Next matches names against XMLTV ids and display-names automatically
- Burst cache / instant join — a second viewer tuning an already-playing channel gets video immediately instead of
waiting for the next file segment
- Mid-stream failover survival — viewers stay connected when a stream fails over to a backup channel; upstream
dropped them
- Slow-client isolation — one stalled client gets disconnected instead of degrading the stream for everyone
- Unit tests + CI — the streaming engine has a test suite and every push runs vet/build/test; upstream had one parser test and release-only CI
Threadfin Next, is a fork of Threadfin that Fable 5 has fully optimized and updated with the following improvements:
⚡ New streaming engine — zero polling, instant channel zaps
🔁 Self-healing streams + auto-failover
📺 EPG maps channels with no tvg-id
🎨 New dark UI + live dashboard
🚀 Auto-updates
See replies for more info!
#iptv #streaming #plex #plexmediaserver #selfhosted
https://t.co/qkmRT3c0r1
Local models are getting good, but the real unlock is the agent layer around them.
Delyx is my attempt at that: model routing, code mode, sandbox execution evidence, source receipts, approvals, memory, autonomy checks, and a desktop/mobile shell that shows what happened instead of hiding it.
Small models. Bigger results. Less guessing. 🧠
Been building this one quietly for a while.
Delyx — a local-first desktop AI agent. runs your own model (Ollama, LM Studio, llama.cpp), can touch files, run commands, do grounded research. every risky action hits a visible approval queue first.
Tauri + Rust under the hood. no daemons, no phone bridges, no surprises.
https://t.co/3rra59iaqH
#LocalLLM #Ollama #BuildInPublic #AIAgents #SelfHosted
@LAGovJeffLandry It wasn't really a pay raise. It was just adding $200 ish on an already existing $1k stipend. Give a real pay raise and performance scale for teachers, and try again.
@7trader13 Goals once I get to coding.
Registered protected_source_roots
path_guard module every write
Containment check before any write
Adversarial tests
Trace marker when guard fires
Edit as diff, not rewrite
Compiler/linter as mando verifier in loop
And other stuff when I get there
@7trader13 Most agent safety work checks what the model says. Not what the system is allowed to do. Refusal + grounding + verification ≠ a filesystem boundary.
Once I get dialed in a bit more, I'll be greatly refining the coding systems. This is one of my first things on the list.