Search with Exa, Ceramic, Linkup, Tavily, Serper, OpenAI, Perplexity, or DeepSeek and return page URLs, titles, and descriptions in one response format. Authenticate with an API key that grants websearch.execute and send Content-Type: application/json.

Provider and credentials

Each request uses the selected provider. Exa uses auto search, Linkup uses standard depth, Tavily uses basic depth, and Perplexity and DeepSeek use their standard Search APIs. DeepSeek accepts queries of at most 128 characters. OpenAI runs one forced web_search tool call on gpt-5.6-luna through the Responses API and returns the raw search results.

Credentials are selected from the authenticated workspace. A configured provider integration takes precedence over ORQ-managed credentials: the default integration is used, or the first configured integration if no default is set. Configure BYOK in the workspace integrations settings. Invalid integration credentials cause an error.

Billing

Managed searches use the same credits as the AI Gateway and charge the provider cost plus US$0.001 per search (US$1 per 1,000 searches). BYOK searches consume no ORQ credits and add no ORQ markup; the provider bills the integration owner directly.

Exa's published auto rate is US$7 per 1,000 requests for up to 10 results, or US$8 per 1,000 including the ORQ markup. Exa charges use the actual cost reported by the provider.

Perplexity's published Search API rate is US$5 per 1,000 requests and DeepSeek's is US$0.75 per 1,000 requests. OpenAI web search costs US$10 per 1,000 tool calls plus the gpt-5.6-luna tokens of the Responses call, computed from the usage reported by OpenAI, so each search is typically US$0.012 to US$0.015 before the ORQ markup.

A successful managed search is billable even if an output guardrail or output redaction failure prevents the results from being returned.

Policies and observability

PII Redaction is the only supported request plugin. Workspace PII settings and matching guardrail rules also apply. Query redaction runs before input guardrails and the provider call; result redaction runs before output guardrails and the response.

Search spans and metrics record the provider, latency, result count, and cost. Use the x-orq-trace-id response header to find the trace.