---

name: parallel-ai-search description: Search the live web and extract clean, LLM-ready excerpts/markdown from URLs (including PDFs and JS-heavy pages) using Parallel Search + Extract APIs. Use for up-to-date research, domain/date-scoped sourcing (include_domains/after_date), and turning specific URLs into citeable text. compatibility: Requires Node.js 18+ (global fetch), network access to https://api.parallel.ai, and the PARALLEL_API_KEY environment variable. metadata: author: "openclaw" version: "1.1.0" homepage: "https://docs.parallel.ai/search/search-quickstart" openclaw: "{"emoji":"🔎","primaryEnv":"PARALLEL_API_KEY"}"

Parallel AI Search

Use this skill to run web research through Parallel Search (ranked, LLM-optimised excerpts) and Parallel Extract (clean markdown from specific URLs, including JS-heavy pages and PDFs).

This skill follows the Agent Skills format: keep SKILL.md focused, and load extra details from references/ as needed.

When to use this skill

Use this skill when the user needs any of the following:

• Up-to-date web research (“look this up”, “find the latest”, “what changed recently”).
• Source control (“only use official docs”, “only these domains”, “after 2025-01-01”).
• Readable extracts from URLs (turn a URL/PDF into clean text/excerpts suitable for quoting/citing).
• Repeatable research loops (search → shortlist → extract → answer with citations).

Preconditions

• PARALLEL_API_KEY must be available in the environment.
• Node.js 18+ is required (the scripts rely on the built-in fetch).

OpenClaw-specific setup notes are in references/openclaw-config.md.

Available scripts

Run scripts using relative paths from the skill root (e.g. node scripts/parallel-search.mjs ...).

• scripts/parallel-search.mjs — Calls Parallel Search (POST /v1beta/search) to discover sources.
• scripts/parallel-extract.mjs — Calls Parallel Extract (POST /v1beta/extract) to extract clean excerpts/markdown from URLs.
• scripts/parallel-search-extract.mjs — Convenience pipeline: search then extract the top N results.

Tip: each script supports --help, --dry-run, and JSON output by default.

Workflow (recommended)

1) Write an objective + queries

• Objective: 1–3 sentences describing the question, preferred source types, and any freshness constraints.
• Queries: 3–8 keyword queries including synonyms, version numbers, dates, or exact error strings.

If you’re unsure, use references/prompting.md templates.

2) Search (discover)

node scripts/parallel-search.mjs \
  --objective "Find official documentation explaining how X works. Prefer sources after 2025-01-01." \
  --query "X official documentation" \
  --query "X changelog 2025" \
  --max-results 8

Then inspect results[].url, results[].title, and results[].publish_date (if present) and pick the best sources.

3) Extract (read)

Extract only the URLs you actually need:

node scripts/parallel-extract.mjs \
  --url "https://example.com/docs/x" \
  --objective "How does X work? Include the most important constraints." \
  --excerpts \
  --no-full-content

Notes:

• Extract supports up to 10 URLs per request; the script auto-batches if you pass more.
• Prefer --excerpts unless you truly need full content.

4) Answer (with citations)

• Prefer official/primary sources when possible.
• Quote/paraphrase only the extracted text you need.
• Include URL + publish date (when present) for transparency.
• If sources disagree, report both and explain why.

High-signal recipes

Recipe A: Domain-scoped research (official-only)

node scripts/parallel-search.mjs \
  --objective "Answer the question using official sources only." \
  --query "X authentication guide" \
  --include-domain "docs.vendor.com" \
  --include-domain "github.com" \
  --max-results 10

Recipe B: Freshness constrained

node scripts/parallel-search.mjs \
  --objective "I need the latest info; prefer sources after 2026-01-01." \
  --query "X release notes" \
  --after-date "2026-01-01" \
  --fetch-max-age-seconds 3600

Recipe C: One command (search → extract top 3)

node scripts/parallel-search-extract.mjs \
  --objective "Find the latest guidance on Y and extract citeable passages." \
  --query "Y documentation" \
  --query "Y 2026 update" \
  --max-results 8 \
  --top 3 \
  --excerpts

Troubleshooting

Missing API key / auth failures

• Symptom: errors mentioning missing PARALLEL_API_KEY or HTTP 401/403.
• Fix: set PARALLEL_API_KEY in the environment. For OpenClaw, see references/openclaw-config.md.

No good results

• Add or refine queries (include synonyms, product names, dates, or exact error messages).
• Add --include-domain to constrain sources to known-good domains.
• Add/adjust --after-date or --fetch-max-age-seconds for freshness.

Timeouts / slow pages during extract

• Use --fetch-timeout-seconds to raise the API-side crawl timeout.
• If you need fresh crawls, set --fetch-max-age-seconds (min 600 for extract).
• If cached content is acceptable, avoid --disable-cache-fallback.

References (load on demand)

• references/parallel-api.md — Compact field/shape reference for Search/Extract requests and responses.
• references/prompting.md — Objective + query templates and research patterns.
• references/openclaw-config.md — OpenClaw config + sandbox environment notes.