I run a self-hosted fanfiction archive, FicHub, built entirely in Rust: an Axum/PostgreSQL server that scrapes and saves stories from 107+ sites (AO3, FanFiction.net, RoyalRoad, forums…) and exports them as EPUB/MOBI/PDF/TXT.

The server (source)

• 107+ scraper adapters, full FanFicFare parity, site cache so fics survive the source disappearing
• Boolean + fielded search, main-character-attribute search ("dark Harry Potter starring Harry"), search inside fic bodies, natural-language "Ask the Archive"
• Bookmarking, 5-star ratings, reviews, threaded comments, reading lists, update feeds
• Personalized recommendations: co-occurrence, decay, embeddings, matrix factorization, hybrid, author/tag-graph, Markov, clusters, bandit — pluggable strategies
• A community roadmap with consensus voting (semantic clustering + Elo-style ranking)
• Self-healing scrape pipeline: structural failures get diagnosed by an on-the-fly LLM agent, validated, and used to complete the export; transient failures go into a pending-request queue that replays on success

The bot workspace (source)

The interesting engineering bit: instead of writing N bots, I wrote one platform-neutral core, archivist-core, and made every adapter a thin translation layer:

Discord ─┐
Matrix ──┤
Telegram─┼──► archivist-core ──► FicHub REST API
Slack ───┤        │                  + Redis (shared)
IRC ─────┤   do_* → PlatformMessage
Fediverse┘
CLI/TUI ─┘

• archivist-core has zero platform SDKs — no discord-rs, no teloxide. Every command is a do_* function returning a platform-neutral PlatformMessage IR (Text/Rich/File/Ephemeral).
• New commands land once in the core; every platform gets them for free. Adapters only translate platform events in and render the IR back out.
• Shared Redis for pagination cache, token store, cross-platform rate limiting.

The CLI (source) — fic-archivist

• search, quote, download (batch: --infile, --format epub,mobi,pdf,html, --out-dir, --force), body-search, bookmarks, recs, ask, link/updates/status, forum subcommands
• Colored TTY or JSON output; a ratatui three-pane TUI (async worker over tokio::sync::mpsc — state machine testable without a terminal); rustyline REPL with intent classification

cargo install fic-archivist --version 0.2.0

Testing — 256 tests across 8 crates, all offline (no Redis/FicHub/Ollama needed to run the suite).

Published crates — fic-archivist, fanfic-archivist (Discord), archivist-core, forum-core, fanfic-scrapers on crates.io. Source on Forgejo, AGPL-3.0-or-later.

The core + thin-adapters split is the piece I think is reusable beyond fanfiction — if you're building a multi-platform bot, it's worth stealing. Happy to answer questions about the adapter pattern, the scraper architecture, or the self-healing agent pipeline.