Skip to content

Open-source Python library for AI browser agents, plus Browser Use Cloud: stealth browsers and a hosted agent API.

Record SDK runtime bridge proof

Magnus Müller2d81a1a2a2f9
diff --git a/RUST_AGENT_INTEGRATION_FEATURES.md b/RUST_AGENT_INTEGRATION_FEATURES.md
index f45d3a6d6..dfc1ea95c 100644
--- a/RUST_AGENT_INTEGRATION_FEATURES.md
+++ b/RUST_AGENT_INTEGRATION_FEATURES.md
@@ -31,11 +31,16 @@ This branch keeps the Python `Agent` unchanged unless callers explicitly import
so a provider request that never returns response headers, or a stream that
opens and then sends no SSE bytes, becomes a retryable transport error instead
of holding a GitHub eval runner indefinitely after `model.turn.request`.
+ - Terminal SDK `agent.run_task` now drives the runtime on a multi-thread Tokio
+ runtime, matching the live model transport bridge's `block_in_place` usage.
+ This prevents SDK evals from stalling after `model.turn.request` before the
+ configured response-open/stream-idle timeout can make progress.
## Current Proof
- terminal `cargo check -p browser-use-cli`
- terminal `cargo test -p browser-use-cli sdk_ -- --nocapture`
+- terminal `cargo test -p browser-use-cli sdk_run_runtime_supports_model_transport_blocking_bridge -- --nocapture`
- terminal `cargo test -p browser-use-cli sdk_json_rpc_agent_run_task_executes_fake_backend_with_normalized_history -- --nocapture`
- terminal `cargo test -p browser-use-llm stream_ -- --nocapture`
- terminal `cargo test -p browser-use-cli sdk_provider_run_config_maps_browser_use_options_to_rust_core -- --nocapture`
diff --git a/browser_use/rust/service.py b/browser_use/rust/service.py
index ed30b4fa0..2765dbf27 100644
--- a/browser_use/rust/service.py
+++ b/browser_use/rust/service.py
@@ -4758,11 +4758,12 @@ class Agent(Generic[Context, AgentStructuredOutput]):
progress_task: asyncio.Task[Any] | None = None
if isinstance(sdk, RustSdkClient):
self.logger.info(
- 'Rust SDK %s starting: max_steps=%s browser_mode=%s browser_id=%s',
+ 'Rust SDK %s starting: max_steps=%s browser_mode=%s browser_id=%s llm_timeout=%s',
method,
params.get('max_steps'),
params.get('browser_mode'),
params.get('browser_id') or '<new>',
+ params.get('llm', {}).get('timeout'),
)
progress_task = asyncio.create_task(self._log_sdk_progress(sdk))
try: