AWS bedrock hardening / native web search

we need to check on this. context: what about toolResult, shouldn't we handle that as well to make tool calling work properly along with anthropic provider which supports that? Not for normal client-side tool calling. In our Bedrock flow, tool calling works through toolUse, not toolResult: outbound tool requests from the model are parsed from contentBlockStart / contentBlockDelta with toolUse in extract-tool-calls.ts non-streaming Bedrock tool calls are parsed from block.toolUse in parse-provider-response.ts when the client sends tool outputs back, we already encode them as Bedrock toolResult blocks in prepare-request-body.ts So ignoring streamed delta.toolResult does not break standard OpenAI-style tool calling. toolResult is the return leg after a tool has run, not the model’s request to call a tool. Where you are right: if we want to support Bedrock-managed/server-side tools and expose their streamed results meaningfully, then toolResult needs explicit handling but that should be a deliberate feature mapping, similar to our Anthropic-specific web_search_tool_result handling, not a generic “treat this like tool calling” patch So: for normal Bedrock tool calling: current behavior is fine for Bedrock server-tool results: we likely need a separate implementation Docs: Bedrock tool use overview: https://docs.aws.amazon.com/bedrock/latest/userguide/tool-use.html Converse / content blocks: https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_ContentBlock.html