Method issueRequest

app/src/AI/Conversation.cpp:1079–1125

* @brief Issues a fresh request to the active provider with the current history. */

 * @brief Issues a fresh request to the active provider with the current history.
 */
void AI::Conversation::issueRequest()
{
  Q_ASSERT(m_provider);
  Q_ASSERT(m_dispatcher);

  pruneHistory();

  reconcileHistoryToolPairs();

  ageHistoryToolResults();

  beginAssistantMessage();

  if (m_provider->capabilities().slowFirstToken)
    m_assistantThinking = tr("Waiting for %1 to respond. Loading the model and processing "
                             "the prompt can take a while on local hardware...")
                            .arg(m_provider->displayName());
  else
    m_assistantThinking = tr("Sending request to %1...").arg(m_provider->displayName());

  m_thinkingIsSynthetic = true;
  if (m_assistantIndex >= 0 && m_assistantIndex < m_uiMessages.size()) {
    auto map = m_uiMessages.at(m_assistantIndex).toMap();
    map.insert(QStringLiteral("thinking"), m_assistantThinking);
    m_uiMessages[m_assistantIndex] = map;
    Q_EMIT messagesChanged();
  }

  const auto tools = dispatcherTools();
  m_reply          = m_provider->sendMessage(budgetedHistory(tools), tools, m_summaryForced);
  if (!m_reply) {
    setLastError(tr("Provider returned no reply"));
    Q_EMIT errorOccurred(m_lastError);
    setBusy(false);
    return;
  }

  connect(m_reply, &Reply::partialText, this, &Conversation::onPartialText);
  connect(m_reply, &Reply::partialThinking, this, &Conversation::onPartialThinking);
  connect(m_reply, &Reply::thinkingBlockFinished, this, &Conversation::onThinkingBlockFinished);
  connect(m_reply, &Reply::toolCallRequested, this, &Conversation::onToolCallRequested);
  connect(m_reply, &Reply::finished, this, &Conversation::onReplyFinished);
  connect(m_reply, &Reply::errorOccurred, this, &Conversation::onReplyError);
  connect(m_reply, &Reply::cacheStatsAvailable, this, [](int read, int created) {
    Assistant::instance().reportCacheStats(read, created);
  });
}

/**
 * @brief Returns the ordered, deduped tool_use ids declared by an assistant message.