Fully migrate off Gemini: shared Ollama client (gemma4:26b @ shadow-death), health reports llm backend, drop @google/genai

src/server/llmClient.ts

@@ -0,0 +1,38 @@
+// Shared local-LLM client — Ollama on shadow-death (.128).
+// Single source of truth for host/model across the whole app.
+export const OLLAMA_HOST = process.env.OLLAMA_HOST || 'http://10.30.20.128:11434';
+export const OLLAMA_MODEL = process.env.OLLAMA_MODEL || 'gemma4:26b';
+
+export interface LocalModelOptions {
+  formatJson?: boolean;
+  numPredict?: number;
+  temperature?: number;
+}
+
+export async function callLocalModel(prompt: string, opts: LocalModelOptions = {}): Promise<string> {
+  const body: Record<string, any> = {
+    model: OLLAMA_MODEL,
+    messages: [{ role: 'user', content: prompt }],
+    stream: false,
+    keep_alive: '30m',
+    options: {
+      num_predict: opts.numPredict ?? 4096,
+      temperature: opts.temperature ?? 0.2,
+    },
+  };
+  if (opts.formatJson) body.format = 'json';
+
+  const res = await fetch(`${OLLAMA_HOST}/api/chat`, {
+    method: 'POST',
+    headers: { 'Content-Type': 'application/json' },
+    body: JSON.stringify(body),
+    signal: AbortSignal.timeout(180000),
+  });
+  if (!res.ok) {
+    throw new Error(`Ollama HTTP ${res.status} (${res.statusText})`);
+  }
+  const data = await res.json();
+  let content = data?.message?.content || '';
+  if (!content) content = data?.message?.thinking || '';
+  return (content || '').trim();
+}