Web Search

OpenMarketing’s web search tool retrieves current information for market research, customer evidence, category changes, pricing, distribution, regulations, and platform behavior. It supports multiple simultaneous queries, domain filtering, recency filtering, and full-page retrieval.

Routing modes

Web search comes from the core pi-web-access package. It supports three backends. You can pick one or let it choose automatically:

Mode	Description
auto	Default. Try Perplexity, then Exa, then the Gemini API, then Gemini in the browser
perplexity	Use Perplexity for all web searches
exa	Use Exa for all web searches
gemini	Use the Gemini API or Gemini in the browser

Default behavior

Without API keys, auto falls through to Gemini in the browser, which uses a signed-in Chromium profile. This works where a Chromium-based browser is installed and signed in to a Google account.

For headless environments, CI pipelines, or servers without a browser, configure an API key for Perplexity, Exa, or Gemini.

Configuration

Check the current search configuration:

openmarketing search status

Set a provider and save its key:

openmarketing search set perplexity <api-key>
openmarketing search set exa <api-key>
openmarketing search set gemini <api-key>

openmarketing search clear resets the provider to auto and keeps saved keys.

The settings live in ~/.openmarketing/agent/web-search.json:

{
  "route": "auto",
  "perplexityApiKey": "pplx-...",
  "exaApiKey": "...",
  "geminiApiKey": "AIza..."
}

Search features

The web search tool supports several capabilities that the researcher agent leverages automatically:

• Multiple queries — Send 2-4 varied-angle queries simultaneously for broader coverage of a topic
• Domain filtering — Restrict results to official product sites, documentation, regulatory sources, review platforms, communities, or other relevant domains
• Recency filtering — Filter by date for fast-moving categories, prices, policies, platforms, and competitors
• Full content retrieval — Fetch complete page content for the most important results rather than relying on snippets

When it runs

Web search is used automatically when a workflow needs current external evidence. Discovery snippets are leads, not proof: important claims should be checked against the underlying page, dated, and preserved with a direct URL. Customer reviews and community discussions remain directional samples unless the evidence supports a broader claim.