Doubao Web Search

Problem

When an agent needs live information—current events, docs, release notes, or external facts—the model’s built-in knowledge is not enough. Doubao WebSearch wraps Doubao’s web search behind a simple async task: submit a query, wait for completion, then consume a structured result instead of hand-rolling polling.

How it works

• Submit search: calls Redfox API’s POST /story/api/doubaoSearch/submit with inquiry_text and the fixed source.
• Poll automatically: queries POST /story/api/doubaoSearch/result every 5 seconds until status is completed, for up to 5 minutes.
• Report progress: waits are written to stderr, keeping the final JSON on stdout clean.
• Parse output: after completion, reads Doubao’s answer from data.content and citations from data.result.

Boundaries

• Requires REDFOX_API_KEY; missing credentials should trigger a clear setup hint.
• Search results depend on Doubao’s live availability; failed, timeout, and submit errors are reported explicitly.
• Query wording matters because keywords are extracted from user intent.