From: Nico <redacted>
Date: Sat, 1 Aug 2026 16:03:32 +0000 (-0400)
Subject: cli : persist reasoning_content in chat history (#26362)
X-Git-Tag: upstream/0.0.10438~219
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=c629da565c80b0b17fac6262acdca4d772e745d8;p=pkg%2Fggml%2Fsources%2Fllama.cpp

cli : persist reasoning_content in chat history (#26362)

* cli : persist reasoning_content in chat history

llama-cli collected reasoning from the stream for display but only
stored assistant content in messages, so --reasoning-preserve could
not re-inject prior thoughts on later turns.
---

diff --git a/tools/cli/cli-context.cpp b/tools/cli/cli-context.cpp
index 0de8f6902..3d801b73d 100644
--- a/tools/cli/cli-context.cpp
+++ b/tools/cli/cli-context.cpp
@@ -624,10 +624,14 @@ int cli_context::run() {
         generated_content content;
         generate_completion(content, timings);
 
-        impl->messages.push_back({
+        json assistant_msg = {
             {"role",    "assistant"},
             {"content", content.content}
-        });
+        };
+        if (!content.reasoning.empty()) {
+            assistant_msg["reasoning_content"] = content.reasoning;
+        }
+        impl->messages.push_back(std::move(assistant_msg));
 
         if (output_file) {
             std::string out_content = "Assistant:\n";