Reclio

Netflix-style personalized recommendations for Chillio, powered by your Trakt.

Reclio is a ChillLink Protocol addon server. It reads your Trakt watch history (and what you don't finish), builds a taste profile, and returns 10 personalized rows — five movie + five show pairs — to Chillio every time the app opens.

Trakt history    ─┐
Ratings          ─┤
Watchlist        ─┼─▶  Reclio  ─▶  Chillio (/feeds — 10 rows)
Watch progress   ─┤    │
Watched %        ─┘    ├─ Recombee (collaborative filtering)
                       ├─ Vector embeddings (semantic similarity)
                       ├─ Watch-state machine (incomplete-watch signal)
                       └─ LLM (chat, personality, intent classification)

What you get

• 10 personalized rows — Recommended For You, Because You Watched, Trending, Top Genre You'll Love, and Hidden Gems, each as a movies + shows pair. Drops the v1.4 grab-bag of 22 sections in favor of a tight, repeatable layout.
• Trakt-native + watch-state aware — learns from watches, ratings, watchlist, and from incomplete watches (the v1.5 watch-state machine catches "fell asleep" vs "actively bounced" and signals Recombee accordingly).
• Ask Reclio chat — floating bubble on the dashboard. Ask it anything ("why am I seeing X?", "more thrillers please", "stop recommending horror") and it can actually act on the answer.
• Personality card — iOS Health-style donut chart of your top genres + an LLM-written one-line roast.
• Self-hostable — one docker compose up and you have your own instance with Caddy-managed HTTPS.
• Never 5xx — every external service (Trakt, TMDB, Recombee, LLM, embeddings) can fail independently and /feeds still returns a valid response.

Two ways to use it

1. Public instance — open reclio.p0xl.com, connect Trakt, paste your addon URL into Chillio. No installation.
2. Self-host — clone the repo, fill in four API keys, run docker compose up -d. See Setup.

Where to go next

• Setup — running Reclio locally or in Docker
• Configuration — every env var explained
• Model integrations — picking a chat LLM (Ollama, Claude, OpenAI, OpenRouter)
• Embeddings — what they are, why we use them, how to pick a provider
• Recombee — how the collaborative filtering works
• Watch-state machine — how Reclio learns from incomplete watches
• API reference — /manifest, /feeds, /admin/*
• Adaptive sync — how sync cadence auto-tunes
• Caddy vs Traefik — picking a reverse proxy
• Troubleshooting — common gotchas