chat client clean up and switch to llama.cpp with models folder network sharing

packages/orchestration-service/src/chat/index.js

@@ -109,31 +109,35 @@ async function chatStream(externalId, userMessage, onChunk, options = {} ) {
     let tokenCount = 0;
 
     // 5. Parse SSE chunks
-    for await (const chunk of res.body){
+    // Replace the current SSE parsing block in chatStream:
+    for await (const chunk of res.body) {
         const lines = chunk.toString().split('\n');
 
         for (const line of lines) {
             if (!line.startsWith('data: ')) continue;
             const raw = line.slice(6).trim();
-            if (raw === '[DONE]') continue //stream closed sentinel
+            if (raw === '[DONE]') continue;
 
             try {
                 const data = JSON.parse(raw);
-                if (data.model) model = data.model
 
+                // llama.cpp provider shape: { response, done }
                 if (data.response) {
                     fullText += data.response;
                     onChunk(data.response);
                 }
 
-                if (data.done && data.eval_count !== undefined) {
-                    tokenCount = (data.eval_count || 0) + (data.prompt_eval_count || 0)
-                }
-            } catch {
-                //partial chunk
-                //skip and wait for next
-            }
+                // model comes through on done chunk from inference route
+                if (data.model) model = data.model;
 
+                // token count — inference.js route sends this on the done chunk
+                if (data.done && data.tokenCount !== undefined) {
+                    tokenCount = data.tokenCount;
+                }
+
+            } catch {
+                // partial chunk — skip
+            }
         }
     }