Configuration

All configuration lives in config.yaml. The repo ships a fully annotated config.example.yaml — copy it and edit. Most options are also editable from the Web UI Settings page, which writes back to the same file. The scheduler reloads config.yaml from disk before every scan, so changes take effect on the next cycle without a restart. The one exception is web.port, which requires a restart.

---

Connections

qbittorrent:
  url: http://192.168.1.100:8080
  username: admin
  password: changeme

arrs:
  sonarr:
    enabled: true
    url: http://192.168.1.100:8989/sonarr     # include the base path if you use one
    api_key: your_api_key
  radarr:
    enabled: true
    url: http://192.168.1.100:7878/radarr
    api_key: your_api_key
  lidarr:
    enabled: false
    url: ~
    api_key: ~

Setting	Purpose
qbittorrent.url	qBittorrent Web UI URL including port
qbittorrent.username / password	qBittorrent Web UI credentials
arrs.<app>.enabled	Whether this *arr client is active
arrs.<app>.url	The *arr URL, including any base path (e.g. /radarr)
arrs.<app>.api_key	Found in the *arr under Settings → General

---

Rules

Rules decide what counts as a bad torrent. Each rule watches one qBittorrent category and ties it to one *arr app. See Rules for the full breakdown.

rules:
  - name: "TV Bad Extensions"
    category: "tv-sonarr"
    app: sonarr
    conditions:
      match_mode: any
      bad_extensions: [".exe", ".bat", ".msi", ".scr", ".lnk"]
      min_file_size_mb: ~
      bad_filename_patterns: []

---

Behavior

on_arr_failure: delete        # "delete" or "abort"
poll_interval_seconds: 300
dry_run: false

Setting	Purpose
on_arr_failure	delete removes the torrent from qBittorrent even if the *arr blocklist call fails; abort skips deletion and queues a retry
poll_interval_seconds	How often the scheduler daemon scans (default 300)
dry_run	true logs matches but takes no action — ideal for testing a new rule

---

Retry

retry:
  enabled: true
  max_attempts: 10
  interval_seconds: 600

When an *arr call fails and on_arr_failure is abort, the torrent is queued and retried up to max_attempts times, interval_seconds apart.

---

Notifications

See Notifications for Pushover setup.

notifications:
  pushover:
    enabled: true
    app_token: your_app_token
    user_key: your_user_key
    notify_on: [action, error, startup, dry_run]
    priority: 0

---

Prowlarr

Enables indexer health scoring, auto-reorder, and grab attribution. See Prowlarr Indexer Scoring for the full explanation of each weight.

prowlarr:
  enabled: true
  url: http://192.168.1.100:9696/prowlarr     # include base path if used
  api_key: your_api_key
  base_priority: 10
  reorder_interval_hours: 24
  history_window_days: 90
  min_grabs_before_scoring: 10
  scoring:
    response_time_weight: 0.25
    failure_rate_weight: 0.30
    malicious_weight: 0.20
    grab_success_weight: 0.25
    auth_failure_mult: 3.0
    grab_failure_mult: 2.0
    query_failure_mult: 1.0
    rss_failure_mult: 0.5
    backoff_penalty: 20
    malicious_penalty_per_hit: 10

Setting	Purpose
base_priority	Priority number given to the best-scoring torrent indexer; others count up from here. Keep it above your NZB indexers' priority numbers so they stay on top
reorder_interval_hours	How often the auto-reorder runs
history_window_days	Rolling window for response-time and failure-rate scoring
min_grabs_before_scoring	Indexers with fewer history records than this show N/A instead of a score
scoring.*	Weights and multipliers for each scoring factor — see Prowlarr Indexer Scoring

Note: The scoring block is preserved when you save from the Settings page even though it has no form fields there. Edit those weights in config.yaml directly.

Ollama (AI scoring)

Optionally point Inspectarr at a local Ollama instance to enable AI-powered health scoring. When configured, the LLM receives every indexer's full statistics — query counts, grab counts, failure rates broken out by type (queries, grabs, RSS, auth), response times, and the deterministic health score — and returns its own 0–100 score with reasoning. The AI score replaces the deterministic score when available.

If Ollama is not configured, not reachable, or returns an error, scoring falls back to the deterministic formula silently — AI is never required.

prowlarr:
  ollama:
    url: http://192.168.1.125:11434      # Ollama API base URL
    model: gemma4:latest                  # any model Ollama has pulled
    timeout: 120                          # request timeout in seconds

Setting	Purpose
ollama.url	Base URL of your Ollama instance (the /api/generate endpoint is appended automatically)
ollama.model	Model name exactly as shown by ollama list
ollama.timeout	How long to wait for a response before falling back to deterministic scoring (default 120)

Leave the entire ollama: block out (or set url to empty) to disable AI scoring entirely.

---

Web UI

web:
  port: 8585
  scheduler_autostart: false
  auth:
    enabled: false
    username: admin
    password: changeme

Setting	Purpose
web.port	Web UI port (default 8585). Changing this requires a restart
web.scheduler_autostart	true starts the scheduler automatically on launch
web.auth.*	HTTP basic auth — see Authentication

---

Advanced

logging:
  log_file: ./data/inspectarr.log.json
  retention_days: 30
  level: INFO              # set to DEBUG for verbose output

state:
  db_file: ./data/inspectarr.db

Setting	Purpose
logging.log_file	Path to the JSON Lines event log
logging.retention_days	Records and log entries older than this are pruned on startup
logging.level	INFO or DEBUG
state.db_file	Path to the SQLite database