garotm - fix the default ollama port expectation.

Author: garotm

backend/src/__tests__/app.test.ts

@@ -28,7 +28,7 @@ describe("createApp HTTP API", () => {
     const app = createApp(db);
     const res = await request(app).get("/api/settings").expect(200);
     expect(res.body.watch_folder).toBeNull();
-    expect(res.body.llm_base_url).toContain("8080");
+    expect(res.body.llm_base_url).toContain("11434");
     expect(res.body.llm_model).toBeTruthy();
   });
 

backend/src/app.ts

@@ -4,6 +4,7 @@ import type { Database } from "better-sqlite3";
 import { z } from "zod";
 import { getSetting, setSetting } from "./db";
 import { ingestFile } from "./ingest";
+import { DEFAULT_LLM_BASE_URL, DEFAULT_LLM_MODEL } from "./llmDefaults";
 
 export function createApp(db: Database) {
   const app = express();
@@ -24,8 +25,8 @@ export function createApp(db: Database) {
   app.get("/api/settings", (_req, res) => {
     const rawWatch = getSetting(db, "watch_folder");
     const watch_folder = rawWatch && rawWatch.length > 0 ? rawWatch : null;
-    const llm_base_url = getSetting(db, "llm_base_url") ?? "http://127.0.0.1:8080/v1";
-    const llm_model = getSetting(db, "llm_model") ?? "local-model";
+    const llm_base_url = getSetting(db, "llm_base_url") ?? DEFAULT_LLM_BASE_URL;
+    const llm_model = getSetting(db, "llm_model") ?? DEFAULT_LLM_MODEL;
     res.json({
       watch_folder,
       llm_base_url,
@@ -47,8 +48,8 @@ export function createApp(db: Database) {
     const watch_folder = wf && wf.length > 0 ? wf : null;
     res.json({
       watch_folder,
-      llm_base_url: getSetting(db, "llm_base_url") ?? "http://127.0.0.1:8080/v1",
-      llm_model: getSetting(db, "llm_model") ?? "local-model",
+      llm_base_url: getSetting(db, "llm_base_url") ?? DEFAULT_LLM_BASE_URL,
+      llm_model: getSetting(db, "llm_model") ?? DEFAULT_LLM_MODEL,
     });
   });
 

backend/src/ingest.ts

@@ -7,6 +7,7 @@ import { extractFromHl7, looksLikeHl7 } from "./pipelines/hl7";
 import { extractTextFromPdf } from "./pipelines/pdf";
 import { getSetting } from "./db";
 import { heuristicSummary, synthesizeClinical } from "./llm";
+import { DEFAULT_LLM_BASE_URL, DEFAULT_LLM_MODEL } from "./llmDefaults";
 
 export type IngestResult = {
   documentId: string;
@@ -91,9 +92,8 @@ export async function ingestFile(db: Database.Database, filePath: string): Promi
   const kind = ext === ".pdf" ? "pdf" : detectKind(normalized, rawText);
   const ctx = await buildContext(kind, normalized, rawText);
 
-  const llmBase =
-    getSetting(db, "llm_base_url") ?? process.env.SIFT_LLM_BASE_URL ?? "http://127.0.0.1:8080/v1";
-  const llmModel = getSetting(db, "llm_model") ?? process.env.SIFT_LLM_MODEL ?? "gpt-oss-20b";
+  const llmBase = getSetting(db, "llm_base_url") ?? process.env.SIFT_LLM_BASE_URL ?? DEFAULT_LLM_BASE_URL;
+  const llmModel = getSetting(db, "llm_model") ?? process.env.SIFT_LLM_MODEL ?? DEFAULT_LLM_MODEL;
 
   let summary: string;
   let confidence = ctx.baseConfidence;

backend/src/llmDefaults.ts

@@ -0,0 +1,4 @@
+/** Defaults aligned with Ollama’s OpenAI-compatible API (`/v1/chat/completions` on port 11434). */
+export const DEFAULT_LLM_BASE_URL = "http://127.0.0.1:11434/v1";
+/** Common tag; change in Settings to match `ollama list` (e.g. `llama3.2`, `mistral`, `qwen2.5:14b`). */
+export const DEFAULT_LLM_MODEL = "llama3.2";

docs/CONFIGURATION.md

@@ -40,7 +40,7 @@ These are exposed via `GET/POST /api/settings` and the UI:
 | `llm_base_url` | Base URL for chat completions (often ends with `/v1`)         |
 | `llm_model`    | Model name passed to the LLM API                              |
 
-Defaults in code if unset: see `backend/src/index.ts` for current defaults.
+Defaults if unset (also used by the UI): **`http://127.0.0.1:11434/v1`** and **`llama3.2`** — aligned with **Ollama**’s OpenAI-compatible API on port **11434** (not 8080). See `backend/src/llmDefaults.ts`. Adjust the model to match `ollama list` on your machine (for example `mistral`, `qwen2.5:14b`).
 
 ## LLM endpoint shape
 

frontend/src/App.tsx

@@ -1,4 +1,7 @@
 import { useCallback, useEffect, useState } from "react";
+
+const OLLAMA_DEFAULT_BASE = "http://127.0.0.1:11434/v1";
+const OLLAMA_DEFAULT_MODEL = "llama3.2";
 import { invoke } from "@tauri-apps/api/core";
 import { open } from "@tauri-apps/plugin-dialog";
 import { api, type DocumentDetail, type DocumentRow, type Health, type Settings } from "./api";
@@ -267,6 +270,12 @@ function SettingsForm(props: {
 }) {
   const [url, setUrl] = useState(props.settings.llm_base_url);
   const [model, setModel] = useState(props.settings.llm_model);
+
+  useEffect(() => {
+    setUrl(props.settings.llm_base_url);
+    setModel(props.settings.llm_model);
+  }, [props.settings.llm_base_url, props.settings.llm_model]);
+
   return (
     <section className="max-w-xl space-y-6 print:hidden">
       <h2 className="text-lg font-semibold text-white">Settings</h2>
@@ -289,30 +298,59 @@ function SettingsForm(props: {
         </div>
       </div>
       <div className="rounded-xl border border-slate-800 bg-slate-900/40 p-5 space-y-3">
-        <label className="block text-sm font-medium text-slate-300">Local LLM (OpenAI-compatible)</label>
-        <p className="text-xs text-slate-500">
-          Point to <code className="text-clinical-teal">llama-server</code>, Ollama, LM Studio, or any
-          OpenAI-compatible URL on this machine.
+        <label className="block text-sm font-medium text-slate-300">Local LLM (OpenAI-compatible API)</label>
+        <p className="text-xs text-slate-500 leading-relaxed">
+          The app does not bundle model weights. It calls your local server using the same shape as OpenAI chat
+          completions (<code className="text-clinical-teal">/v1/chat/completions</code>).{" "}
+          <strong className="text-slate-400">Ollama</strong> listens on{" "}
+          <code className="text-clinical-teal">127.0.0.1:11434</code> by default (not 8080). Use the base URL
+          below and a model name from <code className="text-clinical-teal">ollama list</code>.
         </p>
-        <input
-          className="w-full rounded-lg bg-slate-950 border border-slate-700 px-3 py-2 text-sm"
-          value={url}
-          onChange={(e) => setUrl(e.target.value)}
-          placeholder="http://127.0.0.1:8080/v1"
-        />
-        <input
-          className="w-full rounded-lg bg-slate-950 border border-slate-700 px-3 py-2 text-sm"
-          value={model}
-          onChange={(e) => setModel(e.target.value)}
-          placeholder="model id"
-        />
+        <div className="flex flex-wrap gap-2">
+          <button
+            type="button"
+            className="rounded-md border border-slate-600 px-3 py-1.5 text-xs text-slate-300 hover:bg-slate-800"
+            onClick={() => {
+              setUrl(OLLAMA_DEFAULT_BASE);
+              setModel(OLLAMA_DEFAULT_MODEL);
+            }}
+          >
+            Use Ollama defaults
+          </button>
+        </div>
+        <div>
+          <span className="text-xs text-slate-500">Base URL</span>
+          <input
+            className="mt-1 w-full rounded-lg bg-slate-950 border border-slate-700 px-3 py-2 text-sm font-mono"
+            value={url}
+            onChange={(e) => setUrl(e.target.value)}
+            placeholder={OLLAMA_DEFAULT_BASE}
+            autoComplete="off"
+            spellCheck={false}
+          />
+        </div>
+        <div>
+          <span className="text-xs text-slate-500">Model id</span>
+          <input
+            className="mt-1 w-full rounded-lg bg-slate-950 border border-slate-700 px-3 py-2 text-sm font-mono"
+            value={model}
+            onChange={(e) => setModel(e.target.value)}
+            placeholder={OLLAMA_DEFAULT_MODEL}
+            autoComplete="off"
+            spellCheck={false}
+          />
+        </div>
         <button
           type="button"
           onClick={() => props.onSaveLlm(url, model)}
           className="rounded-lg bg-slate-100 text-slate-900 px-4 py-2 text-sm font-semibold hover:bg-white"
         >
           Save LLM settings
         </button>
+        <p className="text-xs text-slate-600">
+          After saving, drop a new file or re-copy an existing file into the watch folder to regenerate a summary
+          with the LLM. Existing rows keep their stored text until re-ingested.
+        </p>
       </div>
     </section>
   );