pub fn create_react_agent_replaying(
model: Arc<dyn ChatModel>,
tools: ToolRegistry,
source: ReplaySource,
journal: Journal,
) -> Result<Graph>Expand description
Build a prebuilt ReAct agent graph that answers every model and tool call from a recorded journal instead of executing it (exact replay).
The replaying analogue of create_react_agent_with_recording: same
topology, but the nodes wrap model and tools in
crate::replay::ReplayingChatModel / crate::replay::ReplayingTool,
which serve each call from source (matched by sequence + canonical
request hash) and re-journal it into journal. The wrapped model and
tools are never invoked — carry them for their identity (effect class,
tool schemas) and pass panic-on-call sentinels to prove the
zero-outbound guarantee. The registry must offer the same tool
identities (name, description, parameter schema) as the recorded run’s:
schema content feeds the model-call request hash.
Build source and journal from an crate::replay::ExactReplay
session (source() / fresh_journal()) and drive the graph via
crate::replay::ExactReplay::run_and_verify; see
examples/react_record_replay.rs.