Web Research

Available Models

• gemini-search — Google Gemini with web search grounding (default)
• perplexity-fast — Perplexity AI, faster (default)
• perplexity — Perplexity AI
• nomnom — NomNom search model

By default, queries both gemini-search and perplexity-fast in parallel.

Requirements

• curl

Authentication

Set an API key in an environment variable (preferred):

export POLLINATIONS_API_KEY="YOUR_KEY"

Or create a local .env file at .claude/skills/web-research/.env:

POLLINATIONS_API_KEY="YOUR_KEY"

If POLLINATIONS_API_KEY is not set, the script will prompt for a key (input hidden).

Quick usage

.claude/skills/web-research/scripts/web-research.sh "What is pollinations.ai?" 

Choose a model:

.claude/skills/web-research/scripts/web-research.sh --model perplexity-fast "Fact-check this claim with sources"

Compare multiple models:

.claude/skills/web-research/scripts/web-research.sh --models gemini-search,perplexity-fast,nomnom "Compare answers"

Run multi-model in parallel:

.claude/skills/web-research/scripts/web-research.sh --models gemini-search,perplexity-fast --parallel "Compare answers"

Before reviewing a build plan

Run two searches first — "<vendor> <capability> <year>" and "<capability> MCP server" — and record the result. A first-party vendor launch in the last two quarters gets its own section ("why not the managed product"). It's the cheapest part of the review and can invalidate the whole plan, so do it before studying the design.

Numbers from PDFs or long documents

A summarizer given a leading prompt ("did X beat Y?") tends to echo the prompt, including invented quotes. Ask neutrally ("list every numeric result table") and, for any number you'll repeat, read the source text or table yourself.

Notes

• Uses https://gen.pollinations.ai/v1/chat/completions
• Sends Authorization: Bearer <key>