Skip to main content

​
What web search does

When Web Search is enabled, Verba can recognize questions that need current or uncertain information, run a search, and use the results while composing the final answer in the verb’s personality. It is useful for news, live status, current public figures, recent releases, prices, and other facts that can change after the model was trained.

​
Search path and fallback

Verba uses LangSearch as its primary search provider. If that path is unavailable, it can fall back to Brave’s public HTML results. When a data-center IP is blocked or rate limited, the Brave fallback can retry through a rotating proxy pool (up to three proxy attempts). The model should receive the search results and answer normally. Internal <tool_call> markup is not intended to appear in a user-facing response.

​
When to use it

  • Current events and status incidents
  • Policies, pricing, schedules, or leadership that changes
  • Recent releases and live product information
  • Questions where an old answer could mislead the user

​
When to skip it

  • Roleplay or storytelling
  • Static lore and worldbuilding
  • Stable facts already stored in Knowledge
  • Latency-sensitive voice conversations
Search adds network and model-planning latency and can increase credit use. Response Streaming can show text sooner after generation starts, but it cannot remove the time needed to perform the search.

​
Troubleshooting

Retry once and report the timestamp, verb ID, and surface. That markup is an internal planning format and should be intercepted before sending.
Ask a narrower question with names, location, and date. For stable subjects, turn Web Search off and put the canonical facts in Knowledge instead.
Search may include a primary attempt and fallbacks. Disable it for requests that do not need fresh information and reduce model context if appropriate.

AI engine

Configure search alongside context, model, and streaming.

Memory and knowledge

Store stable facts instead of searching for them every turn.