Donsetch

DonSeTch CLI

Prefer Pi's web_search, source_check, and fetch_content tools for ordinary web research. Use DonSeTch through bash when you need site crawling, bot-wall escalation, browser actions, focused extraction, or cheap change/probe checks.

Inspect available options

donsetch --help
donsetch fetch --help
donsetch search --help
donsetch crawl --help

Search

donsetch search "query"
donsetch search "query" --intent code --max-results 10
donsetch search "query" --json

Search finds candidate URLs; fetch the best result afterward. Treat weak=true or low-consensus results cautiously.

Fetch one or several pages

donsetch fetch 'https://example.com'
donsetch fetch 'https://example.com/long-doc' --focus 'error handling'
donsetch fetch 'https://example.com/long-doc' --toc
donsetch fetch 'https://example.com/long-doc' --section 'Authentication'
donsetch fetch 'https://example.com' --must-contain 'required phrase'
donsetch fetch 'https://example.com' --since-last
donsetch fetch 'https://a.example/x' 'https://b.example/y' --budget-tokens 2000

Choose the cheapest useful mode:

• Verification only: --must-contain.
• Unknown long-page structure: --toc, then --section.
• Known topic: --focus.
• Changed since last read: --since-last.
• Full article pagination: --stitch.
• Structured parsing: --json and process with jq.

For JavaScript-only pages, pass deterministic browser actions:

donsetch fetch 'https://example.com/search' \
  --actions '[{"do":"type","selector":"input[q]","text":"query"},{"do":"press","key":"Enter"},{"do":"wait_text","text":"results"}]'

Crawl a site

donsetch crawl 'https://docs.example.com' --mode map
donsetch crawl 'https://docs.example.com' --topic 'authentication' --max-pages 20
donsetch crawl 'https://docs.example.com' --topic 'API reference' --max-pages 30 --deadline 300
donsetch crawl 'https://docs.example.com' --resume '<resume-token>'

Always set --topic for a focused crawl. Start with --mode map when site structure is unknown. Bound work with --max-pages, --max-chars, and --deadline. Resume partial crawls instead of restarting.

Operational rules

• Quote URLs and user-provided arguments.
• Use --json when downstream code must parse results.
• Redirect large output to a temporary file and inspect bounded slices rather than flooding model context.
• Respect robots.txt; do not use --no-robots unless the user explicitly authorizes it.
• Exit code 2 is transient and may be retried; exit code 3 indicates a wall and usually requires another source or fetch strategy.