Module turn 

The turn driver: keeps asking Azure while a round answers itself with tool calls, and ends on an answer, an error, a suspension, or the round budget.

Modules

cancellation 🔒

Keeps a turn that the client abandoned from losing what it had already streamed.

round 🔒

One tool-call round: running the calls the server answers, storing each beside its output, and handing the round’s items on to the next request.

text_response 🔒

Streaming a text answer to the learner, and handing it to the turn once Azure has finished it.

Enums

TurnStart 🔒

What starts a turn: a new message from the user, or one resuming after the client answered a tool call the previous turn suspended on.

Constants

MAX_TOOL_CALL_ROUNDS_PER_TURN 🔒

How many LLM requests one turn may make, bounding a model that keeps calling tools instead of answering.

Functions

answer_tool_call_and_resume_stream

Records a client’s answer to a tool call the turn suspended on, and continues that turn once nothing else is outstanding.

begin_turn 🔒

Shared preamble of both ways a turn can begin: acquire a connection, repair any tool call a dead turn of this conversation left unanswered, build the request the turn runs with, and hand off to stream_turn — or, on a resume that is still waiting on another call, return the ChatbotChatStreamEvent::Suspended stream without ever reaching it.

recover_and_summarize 🔒

Builds the wire event for a round-ending error, folding in the input summary and response ids every call site of recover_from_round_error otherwise repeats.

recover_from_round_error 🔒

What a round that ended in error becomes: logs it, answers whatever tool call the turn left without an output, and either returns the wire event for an error the turn survives or the original error for one that ends it.

report_stream_failure 🔒

Records a failure that killed a turn in error_variants/error_occurrences.

send_chat_request_and_parse_stream

Starts a turn for a new user message, and streams its NDJSON events to the client.

stream_failure_details 🔒

What every chatbot stream failure records beyond its message, so one query finds them all and each row says which turn it belongs to.

stream_turn 🔒

Runs the request rounds of one turn against the LLM and streams its events as NDJSON.