Parallel Web Search

Web Search

Search the web for: $ARGUMENTS

Command

Choose a short, descriptive filename based on the query (e.g., ai-chip-news, react-vs-vue). Use lowercase with hyphens, no spaces. Substitute it into the command inline — $FILENAME and <keyword> below are placeholders, not shell variables; do not copy them verbatim.

parallel-cli search "$ARGUMENTS" -q "<keyword1>" -q "<keyword2>" --json --max-results 10 --excerpt-max-chars-total 27000 -o "/tmp/$FILENAME.json"

Concrete example for a query about React 19:

parallel-cli search "latest React 19 features and adoption" -q "React 19" -q "concurrent rendering" --json --max-results 10 --excerpt-max-chars-total 27000 -o "/tmp/react-19-features.json"

The first argument is the objective — a natural language description of what you’re looking for. It replaces multiple keyword searches with a single call for broad or complex queries. Add -q flags for specific keyword queries to supplement the objective. The -o flag saves the full results to a JSON file for follow-up questions.

Options if needed:

• --after-date YYYY-MM-DD for time-sensitive queries
• --include-domains domain1.com,domain2.com to limit to specific sources
• --exclude-domains domain.com to filter out noisy sources
• --mode turbo for simple fact lookups where speed and cost matter most (p50 ~200ms, lowest cost). English and Japanese queries only
• --mode advanced for harder questions (multi-step, agentic search). Default basic is right for almost everything; escalate to advanced only when basic results are insufficient, and drop to turbo for high-volume simple lookups
• --location us (ISO 3166-1 alpha-2) for geo-targeted results

Parsing results

Do not set max_output_tokens on the command execution — the output is already bounded by --max-results and --excerpt-max-chars-total. Capping output tokens will truncate the JSON and break parsing.

Prefer reading from the saved -o file, not stdout. Even bounded output regularly exceeds harness stdout limits and gets truncated. Read /tmp/$FILENAME.json for the authoritative payload. For each result, extract:

• title, url, publish_date
• Useful content from excerpts (skip navigation noise like menus, footers, “Skip to content”)

Response format

CRITICAL: Every claim must have an inline citation. Use markdown links like Title (opens in a new tab) pulling only from the JSON output. Never invent or guess URLs.

Synthesize a response that:

• Leads with the key answer/finding
• Includes specific facts, names, numbers, dates
• Cites every fact inline as Source Title (opens in a new tab) — do not leave any claim uncited
• Organizes by theme if multiple topics

End with a Sources section listing every URL referenced:

Sources:
- [Source Title](https://example.com/article) (Feb 2026)
- [Another Source](https://example.com/other) (Jan 2026)

This Sources section is mandatory. Do not omit it.

After the Sources section, mention the output file path (/tmp/$FILENAME.json) so the user knows it’s available for follow-up questions.

Setup

If parallel-cli is not found, install and authenticate:

/parallel:parallel-cli-setup

If parallel-cli search returns 403, tell the user balance is likely required. Offer to run parallel-cli balance get, and if needed ask for explicit confirmation before running parallel-cli balance add <amount_cents>. Then retry the original search command.