VibeAround v0.7.6: host-side web search is here
VibeAround v0.7.6 is out: https://github.com/jazzenchen/VibeAround/releases/tag/v0.7.6
The main update: VibeAround can now give agents web search even when the selected model provider does not expose native server-side search.
How it works:
- VibeAround can replace provider-native `web_search` with a local search runtime.
- Search runs through the git-installed `va-search-tool` plugin.
- Results are normalized and fed back through the API Bridge, so the upstream model gets usable search context without needing native search support.
- Search sources are configured locally in Settings; API keys stay in local settings.
- Supported search source: Exa, Tavily, Grok, and Brave.
Also new in this release:
- Test configured web search sources directly from the desktop Settings flow.
- Inspect search requests and source-separated results in the Bridge recorder.
- Better profile connection coverage around bridge launches and provider test flows.
- Fixes for Codex Desktop API bridge model routing, including fake model IDs such as `GPT-5.5` with DeepSeek.
- Compatibility with older Codex Desktop bridge URL scope shapes.
- Clearer DeepSeek `Content Exists Risk` errors when injected search results trigger upstream moderation.
Packages are available for macOS Apple Silicon, Windows x64, and Linux x64.
Download: https://github.com/jazzenchen/VibeAround/releases/tag/v0.7.6
Website: https://vibearound.ai/


Replies
Interesting approach how does latency compare between native provider search vs the va search tool plugin?
VibeAround
@philip_coleman1 Pretty close in practice. VibeAround runs search sources in parallel, so the added overhead is small: mostly the search provider response time plus a bit of local normalization. For normal coding-agent use, the latency difference vs native provider search is usually not very noticeable.