Architecture

Rust backend (Yggdrasil) ↔ Svelte 5 frontend (crown) over the Elyra Framework's typed MessagePack bridge. The Rust binary embeds the built frontend and serves it from elyra://localhost.

elyra-sja/
├── src/
│   ├── main.rs        # App wiring only (~130 lines)
│   ├── app_paths.rs   # data dir, DB URL, migration staging, retention/vacuum
│   ├── http.rs        # one shared, pooled HTTP client + capped body reads
│   ├── cli.rs         # headless `check` subcommand (no window, no database)
│   ├── mcp.rs         # MCP server on stdio, read-only tools (see docs/mcp.md)
│   ├── background.rs  # launchd agent + `tick`: monitors and scheduled probes while closed
│   ├── seo.rs         # single-page analysis engine (GUI-free): checks + scoring
│   ├── extract.rs     # the page as an engine extracts it: reader view, passages
│   ├── schema.rs      # JSON-LD read for completeness, publisher entity, dates
│   ├── dates.rs       # ISO 8601 / HTTP-date / log timestamps → days (no crate)
│   ├── llmstxt.rs     # llms.txt parser (title, summary, links; lists a URL?)
│   ├── intl.rs        # hreflang / language-tag validation (GUI-free)
│   ├── robots.rs      # robots.txt: every user-agent group, not just `*`
│   ├── botlog.rs      # AI-crawler log analysis engine (GUI-free): per bot, per day
│   ├── iprange.rs     # published crawler IP ranges: fetch shape, CIDR, verdicts
│   ├── citations.rs   # citation probe: ask ChatGPT/Claude/Perplexity/Gemini, judge answers
│   ├── jev.rs         # TypeSafe Jev client: typed judgments (coverage, fidelity, stance)
│   ├── links.rs       # internal-link plan from a crawl: candidates in code, judged by Jev
│   ├── gaps.rs        # "why them, not you": your page vs a cited competitor, judged by Jev
│   ├── voice.rs       # AI share of voice from stored probes; questions to track from a crawl
│   ├── generate.rs    # llms.txt + JSON-LD from crawl data (GUI-free)
│   ├── pagespeed.rs   # Core Web Vitals via PageSpeed Insights
│   ├── crawl.rs       # site crawler (sitemap + BFS, robots, aggregation)
│   ├── fs_tools.rs    # sandboxed AI file tools + atomic, confirmed writes
│   ├── license.rs     # offline license validation (ed25519 signature)
│   ├── gsc.rs         # Search Console OAuth (PKCE + loopback) + API client
│   ├── secrets.rs     # macOS Keychain wrapper
│   └── <commands>     # settings, licensing, projects, reports, trends, crawls,
│                      # botlogs, probes, monitors, ai, export, search_console
├── migrations/        # SQLite migrations (0001–0016; embedded)
├── app/               # Vite + Svelte 5 frontend
│   └── src/
│       ├── App.svelte           # shell, nav, views, scheduler timer
│       ├── bindings.ts          # generated by `rata codegen` — do not edit
│       └── lib/                 # Report, SiteReport, Trends, Monitors, locales,
│                                # Settings, Lock, ProjectDialog, Icon, …
├── scripts/           # icns, dmg, updater keygen/sign
├── vendor/elyra-framework   # the framework, as a git submodule
└── .github/workflows/release.yml

Commands (typed bridge)

Frontend calls Rust through the generated api.* facade. Groups:

• Analysis — analyze_page, analyze_and_save, get_report, list_reports, recent_reports, delete_report.
• Crawl — crawl_site, cancel_crawl, list_crawls, get_crawl, get_crawl_page, delete_crawl.
• Internal links — jev_available, plan_internal_links (progress on the links channel), latest_link_plan.
• AI crawlers — analyze_bot_log, unknown_bot_agents, verify_bot_ips (the log's addresses against the operators' published ranges), bot_visit_trends (visits per crawler per day for a project).
• Citation probe — probe_engines_available, probe_citations, list_citations, delete_citation, citation_summary; competitor comparisons: compare_competitor, list_gap_reports, delete_gap_report; schedules: create_probe_schedule, list_probe_schedules, toggle_probe_schedule, delete_probe_schedule.
• Background runner — background_status, set_background_runner (installs or removes the launchd agent that runs elyra-sja tick), list_run_log (every attempted run with outcome and next attempt).
• Generators — generate_llms_txt, generate_faq_schema, generate_article_schema, save_generated.
• Trends — url_trends, compare_reports, share_of_voice; questions to track from a crawl: suggest_questions.
• Projects — list_projects, create_project, update_project, delete_project.
• Monitors — tick_monitors (also runs due probe schedules), run_monitor_now, list_monitors, create_monitor, toggle_monitor, delete_monitor, list_alerts, unread_alerts, mark_alerts_read.
• AI — run_fix, respond_write.
• Search Console — gsc_status, gsc_connect, gsc_disconnect, gsc_properties.
• Settings & license — get_settings, save_settings, license_status, activate_license.
• Export — open_report.

After changing any #[command] signature or a specta type, re-run rata codegen before touching the frontend.

Events (Rust → frontend)

Pushed over channels the frontend subscribes to:

• crawl — crawl progress (status / per-page / done).
• links — internal-link planning progress (per source page / done).
• ai:fix — AI agent activity (status / tool / write).
• ai:confirm — pending write-confirmation requests.

Database (SQLite)

settings also holds tick_last_run, the unix time of the last background run. Two processes may run monitors and schedules at once — the open app and the tick agent — so both claim a row with UPDATE … SET last_run_at = now WHERE last_run_at = <the value read> before running it; one row affected means "mine".

The reports payload is a PageReport, which since 0.6.0 also carries the extracted reader view (reader), the ranked passages (passages) and the publisher entity (entity), and since 0.8.1 the page's outgoing internal links (links); older payloads load with those defaulted.

Secrets are not in the database — see Security.