Host Web Search
Configure VibeAround's host-side web search fallback for agents and model providers that request provider-side web_search support.
VibeAround can supply host-side web search when an agent asks for provider or server-side web_search but the selected upstream model provider cannot run that tool natively.
This capability is not a normal agent-visible function that the client owns. The API Bridge detects server tool declarations, exposes a temporary host search tool to the upstream model, runs configured search providers from the VibeAround host, then feeds the search results back to the model before returning the final answer to the agent.
Request Flow
The bridge does not search before asking the model. The first model call receives a temporary vibearound_web_search tool. If the model decides it needs current information, it calls that tool. If the model can answer without search, no provider search runs.
Internal Search Rounds
The bridge limits the internal fallback loop to a small number of model/tool rounds. One round means:
- The upstream model returns a
vibearound_web_searchtool call. - VibeAround consumes that tool call.
va-search-toolqueries the enabled search sources.- The bridge appends the tool result and asks the model again.
This is a model-turn limit, not a fixed number of search API calls. A direct final answer uses zero search rounds. A model that searches once and then answers uses one round. The limit prevents a model from repeatedly asking the host to search forever.
Search Sources
Enable sources from Settings > Search. VibeAround currently supports:
- Exa
- Tavily
- Grok/xAI web search
When multiple sources are enabled, va-search-tool searches them in parallel. Successful results are concatenated in configured source order. VibeAround does not deduplicate results and does not rerank across providers. If a provider returns a native score, the normalized result includes it. If the provider does not return a score, VibeAround leaves the score empty instead of inventing one.
Search Defaults
The Search settings include:
| Setting | Meaning |
|---|---|
| Max results per source | Maximum results requested from each enabled source. If Exa and Tavily are both enabled and the value is 5, the model may receive up to 10 results before Grok is considered. |
| Search context size | A low, medium, or high hint that controls how much source content a provider should return. |
Incoming agent requests can still specify their own max_results or search_context_size. When they do not, VibeAround passes the saved defaults to va-search-tool.
Operational Notes
- Search is disabled unless the Search tool is enabled in settings.
- Each enabled source needs its own API key.
- The host search runtime is a supervised local process.
- If no search runtime or provider key is available, the model receives a tool error instead of synthetic results.
- Provider-specific behavior still matters. VibeAround normalizes the shape but does not make all search APIs identical.
Supported AI Providers
Built-in provider profile targets, OpenAI-compatible providers, Anthropic-compatible providers, Gemini-compatible providers, and custom endpoint support.
Messaging Channels for AI Agents
Configure Telegram, Slack, Discord, Feishu/Lark, WeChat, DingTalk, WeCom, and other messaging channels for local AI agent continuation and remote control.