After Brave, the next useful backend is probably something more research/discovery-oriented like Exa or Tavily. I’m intentionally grouping them for now because the main product goal is the same: improve source discovery for harder multi-source questions without changing the core contract of the package.
This is less about replacing ordinary search and more about helping web_explore find better candidates on the first pass. We should still keep search separate from fetch and avoid outsourcing the whole product to a provider’s built-in answer synthesis.
Would treat this as a follow-up to the provider abstraction work. Once that lands, we can evaluate which adapter is the cleaner fit first and add one of them as an optional backend.
After Brave, the next useful backend is probably something more research/discovery-oriented like Exa or Tavily. I’m intentionally grouping them for now because the main product goal is the same: improve source discovery for harder multi-source questions without changing the core contract of the package.
This is less about replacing ordinary search and more about helping web_explore find better candidates on the first pass. We should still keep search separate from fetch and avoid outsourcing the whole product to a provider’s built-in answer synthesis.
Would treat this as a follow-up to the provider abstraction work. Once that lands, we can evaluate which adapter is the cleaner fit first and add one of them as an optional backend.