diff --git a/.env.example b/.env.example
index 2ba3877..ac898a9 100644
--- a/.env.example
+++ b/.env.example
@@ -60,7 +60,7 @@ FIRMS_INTERVAL=900
 INGEST_FIRES=1
 
 # ── API keys (managed from the dashboard UI) ──────────────────────────────
-# Keys such as TELEGRAM_TOKEN are stored in the Postgres
+# Keys such as NOUS_API_KEY and TELEGRAM_TOKEN are stored in the Postgres
 # `api_keys` table and managed from the dashboard's "Keys" tab
 # (GET/POST/DELETE /api/keys/{name}) — see app/keystore.py. The FIRMS ingestor
 # currently reads FIRMS_MAP_KEY from .env (above); wiring the Keys-UI store as
@@ -68,15 +68,17 @@ INGEST_FIRES=1
 
 # ── News pipeline (scraper + summarizer, profile `ingest`) ────────────────
 # Hourly: the scraper crawls 257 RSS sources at minute :00 and the summarizer
-# runs the local Ollama map-reduce at minute :05, both writing to the shared osint-db
+# runs the Nous map-reduce at minute :05, both writing to the shared osint-db
 # (tables `articles` + `article_summaries`, created by alembic 003_news).
 # Consume via GET /api/news and GET /api/news/summaries.
-# LLM_URL is the Ollama origin (Pi container cannot use 127.0.0.1 for a laptop
-# GPU). Example: LLM_URL=http://100.91.7.84:11434
-# LLM_URL=
+# NOUS_API_KEY is also (preferably) set in the Keys UI; env is an override.
+# Unset in both env and api_keys = summarizer logs and idles (never crashes).
+NOUS_API_KEY=
+NOUS_BASE_URL=https://inference-api.nousresearch.com/v1
+# Optional LLM knobs
 # SUMMARY_MODEL is an optional override. Leave unset so Settings
 # (app_settings.SUMMARY_MODEL) can reach the summarizer. Code default
-# qwen3:30b-a3b remains after a Postgres miss. Env wins when set.
+# Hermes-4.3-36B remains after a Postgres miss. Env wins when set.
 # SUMMARY_MODEL=
 NEWS_BATCH_SIZE=50
 SUMMARY_WINDOW_HOURS=1
diff --git a/app/keystore.py b/app/keystore.py
index a701fb7..0eb49a4 100644
--- a/app/keystore.py
+++ b/app/keystore.py
@@ -57,6 +57,11 @@ KEY_REGISTRY: dict[str, dict] = {
         "pattern": r"^[0-9a-fA-F]{32}$",
         "example": "32-char hex string (e.g. 5f3c…9a02)",
     },
+    "NOUS_API_KEY": {
+        "description": "Nous Portal API key — hourly news summarizer (inference-api.nousresearch.com).",
+        "pattern": r"^.{16,}$",
+        "example": "key from https://portal.nousresearch.com (API keys page)",
+    },
     "TELEGRAM_TOKEN": {
         "description": "Telegram bot token — push alert notifications to a channel.",
         "pattern": r"^\d{8,10}:[0-9A-Za-z_-]{35}$",
@@ -209,7 +214,7 @@ async def get_api_key(name: str) -> str | None:
     """Read a stored key value — used by ingest services, never by the API.
 
     Returns the raw value (or None when unset) so producers can pass it to
-    external APIs (FIRMS, Telegram, …). Reads live from Postgres, so a
+    external APIs (FIRMS, Nous, Telegram, …). Reads live from Postgres, so a
     key set via the dashboard is picked up on the next poll — no restart.
     """
     await ensure_api_keys_table()
diff --git a/app/main.py b/app/main.py
index df90401..a2e47ee 100644
--- a/app/main.py
+++ b/app/main.py
@@ -605,7 +605,7 @@ async def list_documents(
 async def list_api_keys():
     """List known API keys with set/missing status — masked, never raw.
 
-    Registered keys (FIRMS_MAP_KEY, TELEGRAM_TOKEN)
+    Registered keys (FIRMS_MAP_KEY, NOUS_API_KEY, TELEGRAM_TOKEN)
     are always included. Any extra stored keys are appended.
     """
     return await list_keys()
@@ -638,13 +638,13 @@ async def remove_api_key(name: str):
 
 @app.get("/api/settings", response_model=SettingsOut)
 async def get_settings():
-    """Summarizer model + read-only Ollama URL."""
+    """Summarizer model + read-only Nous base URL."""
     return await get_app_settings()
 
 
 @app.put("/api/settings", response_model=SettingsOut)
 async def put_settings(payload: SettingsIn):
-    """Persist SUMMARY_MODEL. ``llm_url`` is ignored even if sent."""
+    """Persist SUMMARY_MODEL. ``nous_base_url`` is ignored even if sent."""
     try:
         return await set_summary_model(payload.summary_model)
     except SettingsError as exc:
@@ -1084,7 +1084,7 @@ async def camera_hls_segment(camera_id: UUID, u: str = Query(..., min_length=8))
 
 # ── News pipeline (scraper + summarizer) ──────────────────────────────────
 # Backing data for the frontend news panel. Written by the vendored
-# news-scraper (hourly Scrapy crawl) and news-summarizer (hourly Ollama
+# news-scraper (hourly Scrapy crawl) and news-summarizer (hourly Gemini
 # map-reduce) services into the shared osint-db.
 
 @app.get("/api/news", response_model=list[NewsArticleOut])
@@ -1249,7 +1249,7 @@ async def list_news_map(
 
 @app.get("/api/news/models", response_model=NewsModelsOut)
 async def list_news_models():
-    """Ollama model catalog for the summarizer selector. Never 502s."""
+    """Nous model catalog for the summarizer selector. Never 502s."""
     return await list_models()
 
 
diff --git a/app/schemas.py b/app/schemas.py
index 8fd4c68..b61e296 100644
--- a/app/schemas.py
+++ b/app/schemas.py
@@ -316,7 +316,7 @@ class NewsModelsOut(BaseModel):
 
 
 class SettingsIn(BaseModel):
-    """Body for PUT /api/settings. ``llm_url`` is not writable."""
+    """Body for PUT /api/settings. ``nous_base_url`` is not writable."""
 
     summary_model: str = Field(..., min_length=1, max_length=128)
 
@@ -332,10 +332,10 @@ class SettingsIn(BaseModel):
 
 
 class SettingsOut(BaseModel):
-    """Current summarizer settings. ``llm_url`` is read-only."""
+    """Current summarizer settings. ``nous_base_url`` is read-only."""
 
     summary_model: str
-    llm_url: str
+    nous_base_url: str
 
 
 # ─── Aggregations ────────────────────────────────────────────────────────
diff --git a/app/settings_store.py b/app/settings_store.py
index 44434dc..314db42 100644
--- a/app/settings_store.py
+++ b/app/settings_store.py
@@ -17,10 +17,11 @@ from datetime import datetime, timezone
 import httpx
 from sqlalchemy import Column, DateTime, String, Table, Text, func, select, text
 
+import keystore
 from database import async_session, engine, metadata
 
-DEFAULT_LLM_URL = "http://127.0.0.1:11434"
-DEFAULT_SUMMARY_MODEL = "qwen3:30b-a3b"
+DEFAULT_NOUS_BASE_URL = "https://inference-api.nousresearch.com/v1"
+DEFAULT_SUMMARY_MODEL = "Hermes-4.3-36B"
 SETTING_SUMMARY_MODEL = "SUMMARY_MODEL"
 ALLOWED_SETTINGS = frozenset({SETTING_SUMMARY_MODEL})
 MODELS_CACHE_TTL_S = 600.0
@@ -28,7 +29,12 @@ MODELS_TIMEOUT_S = 8.0
 DEFAULT_MODELS_USER_AGENT = "osint-dashboard-news-summarizer"
 
 FALLBACK_MODELS = [
-    "qwen3:30b-a3b",
+    "Hermes-4.3-36B",
+    "Hermes-4-70B",
+    "google/gemini-2.5-flash",
+    "anthropic/claude-haiku-4.5",
+    "openai/gpt-4.1-mini",
+    "x-ai/grok-4",
 ]
 
 app_settings = Table(
@@ -71,14 +77,10 @@ async def ensure_app_settings_table() -> None:
         _ensured = True
 
 
-def llm_url() -> str:
-    """Read-only Ollama origin (env, never writable from the UI)."""
-    raw = (os.getenv("LLM_URL") or "").strip().rstrip("/")
-    for suffix in ("/api/generate", "/api/chat", "/api", "/v1"):
-        if raw.endswith(suffix):
-            raw = raw[: -len(suffix)].rstrip("/")
-            break
-    return raw or DEFAULT_LLM_URL
+def nous_base_url() -> str:
+    """Read-only Nous inference base URL (env, never writable from the UI)."""
+    raw = (os.getenv("NOUS_BASE_URL") or "").strip().rstrip("/")
+    return raw or DEFAULT_NOUS_BASE_URL
 
 
 def _validate_summary_model(value: str) -> str:
@@ -89,7 +91,7 @@ def _validate_summary_model(value: str) -> str:
 
 
 async def get_summary_model() -> str:
-    """Stored SUMMARY_MODEL, else env, else qwen3:30b-a3b."""
+    """Stored SUMMARY_MODEL, else env, else Hermes-4.3-36B."""
     await ensure_app_settings_table()
     async with async_session() as session:
         row = (
@@ -133,7 +135,7 @@ async def set_summary_model(value: str) -> dict:
 async def get_app_settings() -> dict:
     return {
         "summary_model": await get_summary_model(),
-        "llm_url": llm_url(),
+        "nous_base_url": nous_base_url(),
     }
 
 
@@ -144,6 +146,18 @@ def _fallback_payload() -> dict:
     }
 
 
+async def _nous_api_key() -> str | None:
+    """Keystore first, then env. Any lookup failure is treated as missing."""
+    try:
+        stored = await keystore.get_api_key("NOUS_API_KEY")
+    except Exception:
+        stored = None
+    if stored and str(stored).strip():
+        return str(stored).strip()
+    env = (os.getenv("NOUS_API_KEY") or "").strip()
+    return env or None
+
+
 def _models_user_agent() -> str:
     return os.getenv("OSINT_USER_AGENT") or DEFAULT_MODELS_USER_AGENT
 
@@ -160,7 +174,7 @@ def _parse_models_payload(body: object) -> list[dict[str, str]]:
         if isinstance(item, str) and item.strip():
             out.append({"id": item.strip()})
         elif isinstance(item, dict):
-            mid = item.get("id") or item.get("name") or item.get("model")
+            mid = item.get("id") or item.get("name")
             if mid:
                 out.append({"id": str(mid)})
     return out
@@ -172,15 +186,23 @@ async def _http_get(url: str, *, headers: dict[str, str], timeout: float) -> htt
 
 
 async def list_models() -> dict:
-    """Live ``GET {llm_url}/api/tags``. Never 502 — fallback on failure."""
+    """Live ``GET {base}/models`` when a key is present; otherwise curated fallback.
+
+    Never raises to the caller for missing key or upstream failure.
+    """
     global _models_cache
+    key = await _nous_api_key()
+    if not key:
+        return _fallback_payload()
+
     now = time.monotonic()
     hit = _models_cache
     if hit and now - hit[0] < MODELS_CACHE_TTL_S:
         return hit[1]
 
-    url = f"{llm_url()}/api/tags"
+    url = f"{nous_base_url()}/models"
     headers = {
+        "Authorization": f"Bearer {key}",
         "User-Agent": _models_user_agent(),
         "Accept": "application/json",
     }
diff --git a/app/static/index.html b/app/static/index.html
index c4f8190..db4fb01 100644
--- a/app/static/index.html
+++ b/app/static/index.html
@@ -840,7 +840,7 @@
                     <div class="ns-kicker">Executive Summary <span class="ns-badge" id="ns-badge">WAITING</span></div>
                     <div class="ns-time" id="ns-time">—</div>
                     <div class="ns-body" id="ns-body">
-                        <div class="ns-empty">No executive summary yet — the Ollama summarizer is idle until a model is chosen in Settings and <code>LLM_URL</code> is reachable.</div>
+                        <div class="ns-empty">No executive summary yet — the Nous summarizer is idle until <code>NOUS_API_KEY</code> is set on the Pi and a model is chosen in Settings.</div>
                     </div>
                 </div>
 
@@ -968,7 +968,7 @@
         <section class="view" id="view-keys" aria-label="API keys">
             <div class="subview">
                 <h2><span class="tick">⚿</span> API Keys</h2>
-                <p class="sub">Keys used by ingest services (FIRMS, Telegram, AIS, …). Stored in Postgres, never shown again — only status + last 4 chars.</p>
+                <p class="sub">Keys used by ingest services (FIRMS, Nous Portal, Telegram, AIS, …). Stored in Postgres, never shown again — only status + last 4 chars.</p>
                 <div class="key-grid" id="keys-grid"></div>
             </div>
         </section>
@@ -997,11 +997,11 @@
                         <p class="set-note" style="margin-bottom:0.9rem">Takes effect on the next hourly run (:05). No container restart.</p>
                         <div class="set-row">
                             <label>Provider</label>
-                            <span>Local Ollama (LLM_URL)</span>
+                            <span>Nous Portal (inference-api.nousresearch.com)</span>
                         </div>
                         <div class="set-row">
                             <label>API key</label>
-                            <span>none — Ollama on the GPU host</span>
+                            <span>set in API Keys as NOUS_API_KEY (never shown here)</span>
                         </div>
                         <div class="set-row">
                             <label for="set-summary-model">Model</label>
@@ -1370,7 +1370,7 @@ function renderNewsSummary(summaries) {
     const badgeEl = document.getElementById('ns-badge');
     if (!bodyEl) return;
     if (!summaries || !summaries.length) {
-        bodyEl.innerHTML = '<div class="ns-empty">No executive summary yet — the Ollama summarizer is idle until a model is chosen in Settings and <code>LLM_URL</code> is reachable.</div>';
+        bodyEl.innerHTML = '<div class="ns-empty">No executive summary yet — the Nous summarizer is idle until <code>NOUS_API_KEY</code> is set on the Pi and a model is chosen in Settings.</div>';
         timeEl.textContent = '—';
         badgeEl.textContent = 'WAITING';
         return;
@@ -1547,6 +1547,7 @@ function sentimentBadge(label) {
 /* ═══════════════ API KEYS ═══════════════ */
 const KEY_PATTERNS = {
     'FIRMS_MAP_KEY': /^[0-9a-fA-F]{32}$/,
+    'NOUS_API_KEY': /^.{16,}$/,
     'TELEGRAM_TOKEN': /^\d{8,10}:[0-9A-Za-z_-]{35}$/,
 };
 async function loadKeys() {
diff --git a/docker-compose.yml b/docker-compose.yml
index 400c6b8..aa674f7 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -224,7 +224,8 @@ services:
       DB_HOST: db
       DB_PORT: ${DB_PORT:-5432}
       DB_NAME: ${DB_NAME:-osint_data}
-      LLM_URL: ${LLM_URL:-}
+      NOUS_API_KEY: ${NOUS_API_KEY:-}
+      NOUS_BASE_URL: ${NOUS_BASE_URL:-https://inference-api.nousresearch.com/v1}
       SUMMARY_MODEL: ${SUMMARY_MODEL:-}
       OSINT_USER_AGENT: ${OSINT_USER_AGENT:-osint-dashboard-news-summarizer}
       BATCH_SIZE: ${NEWS_BATCH_SIZE:-50}
diff --git a/docs/news.md b/docs/news.md
index 41e5ced..43a39c0 100644
--- a/docs/news.md
+++ b/docs/news.md
@@ -1,12 +1,11 @@
-# News pipeline — scraper + local Ollama summarizer
+# News pipeline — scraper + Nous Portal summarizer
 
 The OSINT dashboard ingests ~257 global news RSS sources hourly and produces
 an English LLM brief plus flagged ticker/map rows. Both services were vendored
 from the upstream `~/Projects/newsPipeline` project and re-integrated here to
 replace the old k8s CronJob choreography with in-compose scheduling against
-the EXISTING osint-db — **no second Postgres**. The LLM is **local Ollama**
-(`LLM_URL`, default `http://127.0.0.1:11434`) — the newsPipeline `local_llm`
-path (`POST /api/generate`). No API key.
+the EXISTING osint-db — **no second Postgres**. The LLM is **Nous Portal**
+(`inference-api.nousresearch.com`) — not Gemini.
 
 ## Architecture
 
@@ -17,7 +16,7 @@ path (`POST /api/generate`). No API key.
 news-scraper   (Scrapy, hourly :00) ──► articles table (osint-db)
         │                                      │
         │                                      ▼
-news-summarizer (Ollama map-reduce, :05) ──► article_summaries + news_items
+news-summarizer (Nous Portal map-reduce, :05) ──► article_summaries + news_items
                                                       │
                                                       ▼
      GET /api/news · /api/news/summaries · /api/news/ticker · /api/news/map
@@ -47,8 +46,8 @@ feeds (`GET /api/news` exact key set is unchanged on purpose).
    (`ON CONFLICT (url) DO NOTHING`).
 2. **Summarizer** — `news/summerizer/run_news_summarizer.py` runs
    `summarizer.py` at :05 past each hour. It reads articles from the last
-   `SUMMARY_WINDOW_HOURS`, map-reduces them through local Ollama
-   (`SUMMARY_MODEL` / Settings, default `qwen3:30b-a3b`), writes the English
+   `SUMMARY_WINDOW_HOURS`, map-reduces them through Nous Portal
+   (`SUMMARY_MODEL` / Settings, default `Hermes-4.3-36B`), writes the English
    brief to `article_summaries` (column `model` is the LLM id), and flagged
    ticker/map rows to `news_items`.
 
@@ -68,16 +67,17 @@ Container startup order doesn't matter.
 
 ## Keys and Settings
 
-- **No API key.** Ollama on the GPU host; the summarizer container calls
-  `LLM_URL` (`POST /api/generate`, `GET /api/tags`).
-- **Idle without Ollama** — if `/api/tags` is down or the chosen model is not
-  pulled, the summarizer logs and idles (never crashes). News intel APIs
-  return `[]`.
+- **`NOUS_API_KEY`** — paste in the dashboard **Keys** UI (`api_keys` /
+  `keystore.KEY_REGISTRY`). Env / `.env` is an **override** (env wins, same
+  as FIRMS). Never returned by any API; never emitted into `index.html`;
+  never proxied from the browser.
+- **Idle without a key** — if env is unset **and** the keystore row is empty,
+  the summarizer logs and idles (never crashes). News intel APIs return `[]`.
 - **Model** — non-secret. Settings UI model selector `PUT /api/settings`
   `{ "summary_model": "…" }` stores `SUMMARY_MODEL` in `app_settings` (1–128
-  chars). `GET /api/settings` echoes `{summary_model, llm_url}`.
-  `llm_url` is read-only. Default `qwen3:30b-a3b`. Live catalog is
-  best-effort `GET /api/news/models` (`GET {LLM_URL}/api/tags`).
+  chars). `GET /api/settings` echoes `{summary_model, nous_base_url}`.
+  `nous_base_url` is read-only. Default `Hermes-4.3-36B`. Live catalog is
+  best-effort `GET /api/news/models`.
 
 ## Endpoints
 
@@ -192,15 +192,15 @@ Never 502s. `{ "source": "live"|"fallback", "models": [{"id": "…"}] }`.
 ### GET /api/settings · PUT /api/settings
 
 ```json
-{ "summary_model": "qwen3:30b-a3b", "llm_url": "http://127.0.0.1:11434" }
+{ "summary_model": "Hermes-4.3-36B", "nous_base_url": "https://inference-api.nousresearch.com/v1" }
 ```
 
-PUT body is `{ "summary_model": "<1–128 char id>" }`. `llm_url` is
+PUT body is `{ "summary_model": "<1–128 char id>" }`. `nous_base_url` is
 ignored even if sent.
 
 ## Reduce JSON contract
 
-Reduce phase (`format: json` on Ollama generate, English only) must be a single
+Reduce phase (`response_format: json_object`, English only) must be a single
 object. Parser (`intel.parse_reduce_json`) strips `<think>…</think>` and
 markdown json fences, then brace-slices:
 
@@ -234,8 +234,9 @@ lands in `article_summaries.summary_text`.
 
 | Var | Default | Notes |
 |---|---|---|
-| `LLM_URL` | `http://127.0.0.1:11434` | Ollama origin. Pi container must point at the GPU host (e.g. Tailscale). Compose leaves this empty so `.env` wins. |
-| `SUMMARY_MODEL` | `qwen3:30b-a3b` | Leave compose unset so Settings → `app_settings.SUMMARY_MODEL` reaches the worker. Env wins when set. |
+| `NOUS_API_KEY` | *(blank)* | **Required for summaries.** Prefer Keys UI; env overrides. Unset in **both** env and `api_keys` = summarizer logs and idles (never crashes); APIs return `[]`. |
+| `NOUS_BASE_URL` | `https://inference-api.nousresearch.com/v1` | Read-only in Settings. |
+| `SUMMARY_MODEL` | `Hermes-4.3-36B` | Compose default. Operator-facing choice is Settings → `app_settings.SUMMARY_MODEL`. |
 | `NEWS_BATCH_SIZE` | `50` | Articles per map-phase batch (compose maps to container `BATCH_SIZE`). |
 | `SUMMARY_WINDOW_HOURS` | `1` | How far back the summarizer looks for new articles. |
 | `INCLUDE_FUTURES` | `0` | Legacy futures-prices coupling (upstream pipeline). OFF for OSINT; set `1` + install `yfinance` to enable. |
@@ -245,14 +246,15 @@ lands in `article_summaries.summary_text`.
 | `NEWS_SUMMARIZE_RUN_ON_START` | `1` | Run one summarize immediately on container start. |
 | `NEWS_SUMMARIZE_FORCE` | `0` | `1` ignores the current-UTC-hour idempotency skip (double-pins on recreate). |
 | `NEWS_LOG_LEVEL` | `INFO` | Scrapy log level. |
-| `OSINT_USER_AGENT` | `osint-dashboard-news-summarizer` | Sent on every outbound Ollama call. |
+| `OSINT_USER_AGENT` | `osint-dashboard-news-summarizer` | Sent on every outbound Nous call. |
 | `TELEGRAM_TOKEN` / `TELEGRAM_CHAT_ID` | *(blank)* | Reserved for the (out-of-scope) Telegram delivery bot. |
 
 DB_* for both services is mapped to the shared osint-db credentials
 (`DB_HOST=db`, same `DB_USER/DB_PASSWORD/DB_NAME` as the rest of the stack).
 
-Ollama generate: `POST {LLM_URL}/api/generate` via `news/summerizer/ollama_client.py`
-(`httpx`, no `ollama` Python SDK). No API key. Reduce uses `format: json`.
+Nous chat: `POST {NOUS_BASE_URL}/chat/completions` via `news/summerizer/nous_client.py`
+(`httpx`, no `openai` SDK). Auth is a Bearer token from `NOUS_API_KEY`.
+No Hermes-4 reasoning system prompt. Reduce uses `json_mode=True`.
 
 ## Prompts
 
@@ -266,7 +268,7 @@ language is gated behind `INCLUDE_FUTURES=1`.
 
 ```bash
 PYTHONPATH=news/summerizer pytest news/summerizer/tests -v
-# intel + ollama_client tests PASS (no network)
+# intel + nous_client tests PASS (no network)
 
 PYTHONPATH=app pytest tests/test_api_news.py \
   tests/test_api_settings.py tests/test_api_live_layers.py -v
@@ -278,17 +280,16 @@ PYTHONPATH=app pytest tests/test_api_news.py \
 
 After deploy / compose rebuild of `news-summarizer` on the Pi:
 
-1. Run Ollama on the GPU host and pull the model (`ollama pull qwen3:30b-a3b`).
-2. Set `LLM_URL` on the Pi to that host (Tailscale IP + `:11434`).
-3. Settings: pick a model → Save → `GET /api/settings` echoes it.
-4. `docker compose --profile ingest logs -f news-summarizer` — next run (or
+1. Keys UI: save `NOUS_API_KEY` → status `****last4`.
+2. Settings: pick a model → Save → `GET /api/settings` echoes it.
+3. `docker compose --profile ingest logs -f news-summarizer` — next run (or
    `NEWS_SUMMARIZE_RUN_ON_START=1` recreate) logs `Processing N articles with <model>`.
-5. `curl -s localhost:8000/api/news/summaries?limit=1` — English `summary_text`, `model` set.
-6. `curl -s localhost:8000/api/news/ticker` — flagged headlines only.
-7. `curl -s localhost:8000/api/news/map` — only rows with lat/lon.
-8. HUD: NEWS ticker scrolls flagged items; map overlay pins popup with location.
-9. Ollama down / model missing → summarizer logs idle, APIs return `[]`, no crash.
+4. `curl -s localhost:8000/api/news/summaries?limit=1` — English `summary_text`, `model` set.
+5. `curl -s localhost:8000/api/news/ticker` — flagged headlines only.
+6. `curl -s localhost:8000/api/news/map` — only rows with lat/lon.
+7. HUD: NEWS ticker scrolls flagged items; map overlay pins popup with location.
+8. Unset key + empty keystore → summarizer logs idle, APIs return `[]`, no crash.
 
-**Operator action after merge:** start Ollama on the laptop GPU, set `LLM_URL`
-in Pi `.env`, pick a model in Settings if the default `qwen3:30b-a3b` is not
-wanted; rebuild `osint-news-summarizer` on the Pi (`pi-app-deploy` / compose).
+**Operator action after merge:** paste a Nous Portal API key in API Keys; pick
+a model in Settings if the default `Hermes-4.3-36B` is not wanted; rebuild
+`osint-news-summarizer` on the Pi (`pi-app-deploy` / compose).
diff --git a/news/summerizer/nous_client.py b/news/summerizer/nous_client.py
new file mode 100644
index 0000000..da0b2c4
--- /dev/null
+++ b/news/summerizer/nous_client.py
@@ -0,0 +1,36 @@
+"""HTTP client for the Nous inference chat completions API."""
+
+from __future__ import annotations
+
+import os
+
+import httpx
+
+_DEFAULT_UA = "osint-dashboard-news-summarizer"
+_DEFAULT_BASE = "https://inference-api.nousresearch.com/v1"
+
+
+def chat(prompt, *, api_key, model, base_url, json_mode=False) -> str:
+    resolved = (base_url or os.environ.get("NOUS_BASE_URL", _DEFAULT_BASE)).rstrip("/")
+    url = f"{resolved}/chat/completions"
+    headers = {
+        "Authorization": f"Bearer {api_key}",
+        "User-Agent": os.environ.get("OSINT_USER_AGENT") or _DEFAULT_UA,
+    }
+    payload = {
+        "model": model,
+        "messages": [{"role": "user", "content": prompt}],
+        "temperature": 0.2,
+        "max_tokens": 4096,
+    }
+    if json_mode:
+        payload["response_format"] = {"type": "json_object"}
+    try:
+        with httpx.Client(timeout=60.0) as client:
+            resp = client.post(url, headers=headers, json=payload)
+        if resp.status_code == 401 or resp.status_code >= 500:
+            return ""
+        data = resp.json()
+        return data["choices"][0]["message"]["content"]
+    except Exception:
+        return ""
diff --git a/news/summerizer/ollama_client.py b/news/summerizer/ollama_client.py
deleted file mode 100644
index 057b437..0000000
--- a/news/summerizer/ollama_client.py
+++ /dev/null
@@ -1,102 +0,0 @@
-"""HTTP client for a local Ollama generate API.
-
-Matches the newsPipeline local_llm summarizer: POST /api/generate,
-GET /api/tags. No API key. ``LLM_URL`` may be a host, ``/api``, or
-``/api/generate`` — we normalize to the origin.
-"""
-
-from __future__ import annotations
-
-import os
-
-import httpx
-
-_DEFAULT_UA = "osint-dashboard-news-summarizer"
-_DEFAULT_BASE = "http://127.0.0.1:11434"
-_GENERATE_TIMEOUT = 300.0
-_TAGS_TIMEOUT = 8.0
-
-_GENERATE_OPTIONS = {
-    "num_predict": 4096,
-    "temperature": 0.6,
-    "top_p": 0.9,
-    "top_k": 40,
-    "num_ctx": 32768,
-    "repeat_penalty": 1.1,
-}
-
-
-def normalize_base(url: str | None) -> str:
-    """Strip path suffixes so LLM_URL variants share one origin."""
-    raw = (url or os.environ.get("LLM_URL", "") or _DEFAULT_BASE).strip()
-    raw = raw.rstrip("/")
-    for suffix in ("/api/generate", "/api/chat", "/api", "/v1"):
-        if raw.endswith(suffix):
-            raw = raw[: -len(suffix)].rstrip("/")
-            break
-    return raw or _DEFAULT_BASE
-
-
-def _headers() -> dict[str, str]:
-    return {
-        "User-Agent": os.environ.get("OSINT_USER_AGENT") or _DEFAULT_UA,
-        "Accept": "application/json",
-    }
-
-
-def chat(prompt: str, *, model: str, base_url: str | None = None, json_mode: bool = False) -> str:
-    """POST /api/generate. Returns response text, or \"\" on any failure."""
-    if not (model or "").strip():
-        return ""
-    origin = normalize_base(base_url)
-    url = f"{origin}/api/generate"
-    payload: dict = {
-        "model": model.strip(),
-        "prompt": prompt,
-        "stream": False,
-        "options": dict(_GENERATE_OPTIONS),
-    }
-    if json_mode:
-        payload["format"] = "json"
-    try:
-        with httpx.Client(timeout=_GENERATE_TIMEOUT) as client:
-            resp = client.post(url, headers=_headers(), json=payload)
-        if resp.status_code == 401 or resp.status_code >= 500:
-            return ""
-        data = resp.json()
-        text = data.get("response") if isinstance(data, dict) else None
-        return text if isinstance(text, str) else ""
-    except Exception:
-        return ""
-
-
-def list_tags(base_url: str | None = None) -> list[str]:
-    """GET /api/tags model names. Empty list on failure."""
-    origin = normalize_base(base_url)
-    url = f"{origin}/api/tags"
-    try:
-        with httpx.Client(timeout=_TAGS_TIMEOUT) as client:
-            resp = client.get(url, headers=_headers())
-        if resp.status_code != 200:
-            return []
-        models = (resp.json() or {}).get("models") or []
-        names: list[str] = []
-        for item in models:
-            if isinstance(item, dict):
-                name = item.get("name") or item.get("model") or ""
-                if name:
-                    names.append(str(name))
-        return names
-    except Exception:
-        return []
-
-
-def ollama_ready(model: str, *, base_url: str | None = None) -> bool:
-    """True when /api/tags is up and ``model`` is present (prefix match)."""
-    if not (model or "").strip():
-        return False
-    names = list_tags(base_url)
-    if not names:
-        return False
-    want = model.strip()
-    return any(n == want or n.startswith(want) for n in names)
diff --git a/news/summerizer/run_news_summarizer.py b/news/summerizer/run_news_summarizer.py
index 483ea9b..91740c6 100644
--- a/news/summerizer/run_news_summarizer.py
+++ b/news/summerizer/run_news_summarizer.py
@@ -1,11 +1,16 @@
 #!/usr/bin/env python3
 """Scheduler loop for the news summarizer — hourly summarize at minute :05.
 
+Replaces the k8s CronJob (`5 * * * *`) with an in-compose loop. Runs once on
+boot (catches up on any articles scraped since the last summary), then fires
+at each :NEWS_SUMMARIZE_MINUTE wall-clock boundary.
+
+The loop is serial, so a slow LLM pass never overlaps the next run.
+
 Env (all optional, 12-factor):
   NEWS_SUMMARIZE_MINUTE        minute of the hour to fire (default 5)
   NEWS_SUMMARIZE_RUN_ON_START  "1" to summarize once immediately on boot (default 1)
-  LLM_URL                      Ollama origin; default http://127.0.0.1:11434
-  SUMMARY_MODEL                Ollama tag (else app_settings)
+  NOUS_API_KEY                 optional in env; Keys UI / api_keys also works
 """
 
 from __future__ import annotations
@@ -41,10 +46,13 @@ def run_summarize() -> None:
 
 
 def main() -> None:
-    llm_url = os.getenv("LLM_URL", "").strip() or "http://127.0.0.1:11434"
+    if not os.getenv("NOUS_API_KEY", "").strip():
+        logger.warning(
+            "NOUS_API_KEY unset in env — will read api_keys on each run; idle if both empty"
+        )
     logger.info(
-        "news summarizer loop starting (minute=%s, run_on_start=%s, llm_url=%s)",
-        MINUTE, RUN_ON_START, llm_url,
+        "news summarizer loop starting (minute=%s, run_on_start=%s)",
+        MINUTE, RUN_ON_START,
     )
     if RUN_ON_START:
         run_summarize()
diff --git a/news/summerizer/summarizer.py b/news/summerizer/summarizer.py
index 196f3b9..ea82a37 100644
--- a/news/summerizer/summarizer.py
+++ b/news/summerizer/summarizer.py
@@ -1,17 +1,19 @@
 #!/usr/bin/env python3
-"""News summarizer — local Ollama map-reduce into brief/ticker/map.
+"""News summarizer — Nous map-reduce of scraped articles into brief/ticker/map.
 
 Reads articles scraped within the last hour from the shared `articles` table,
-maps them with a local Ollama model (per-article English fact blocks), reduces
-to one JSON object (summary_en + ticker + map_items), and stores the brief in
+maps them with Nous (per-article English fact blocks), reduces to one JSON
+object (summary_en + ticker + map_items), and stores the brief in
 `article_summaries` plus flagged rows in `news_items`. Tables live in the
 EXISTING osint-db (alembic 003_news + 005_news_items, idempotent).
 
-LLM is the newsPipeline local_llm path: Ollama POST /api/generate. No API key.
+Everything is env-driven (12-factor). Secrets/config are resolved at the start
+of each summarize_news() — env wins, else api_keys / app_settings:
 
   DB_HOST / DB_NAME / DB_USER / DB_PASSWORD / DB_PORT   PostgreSQL (osint-db)
-  LLM_URL                  Ollama origin (default http://127.0.0.1:11434)
-  SUMMARY_MODEL            Ollama tag (else app_settings; default qwen3:30b-a3b)
+  NOUS_API_KEY             Nous Portal key (else api_keys.name='NOUS_API_KEY')
+  NOUS_BASE_URL            default https://inference-api.nousresearch.com/v1
+  SUMMARY_MODEL            default Hermes-4.3-36B (else app_settings)
   BATCH_SIZE               articles per map-phase batch (default 50)
   SUMMARY_WINDOW_HOURS     look-back window in hours (default 1)
   OSINT_USER_AGENT         default osint-dashboard-news-summarizer
@@ -19,6 +21,11 @@ LLM is the newsPipeline local_llm path: Ollama POST /api/generate. No API key.
   SUMMARY_PROMPT           override reduce-phase prompt (uses {final_input})
   NEWS_SUMMARIZE_FORCE     "1" to ignore the current-UTC-hour idempotency skip
   INCLUDE_FUTURES          "1" to prepend live futures prices (default 0)
+
+The futures/markets coupling from the original pipeline is gated behind
+INCLUDE_FUTURES and OFF by default — it is irrelevant to the OSINT dashboard
+and pulled yfinance into the image. Re-enable by installing yfinance and
+setting INCLUDE_FUTURES=1.
 """
 
 from __future__ import annotations
@@ -30,7 +37,7 @@ from datetime import datetime
 import psycopg2
 
 from intel import parse_reduce_json, select_map, select_ticker
-from ollama_client import chat, ollama_ready
+from nous_client import chat
 
 logging.basicConfig(level=logging.INFO, format="%(asctime)s %(levelname)s %(message)s")
 logger = logging.getLogger("news.summarizer")
@@ -44,8 +51,8 @@ DB_CONFIG = {
     "port": int(os.getenv("DB_PORT", "5432")),
 }
 
-DEFAULT_LLM_URL = "http://127.0.0.1:11434"
-DEFAULT_SUMMARY_MODEL = "qwen3:30b-a3b"
+DEFAULT_NOUS_BASE_URL = "https://inference-api.nousresearch.com/v1"
+DEFAULT_SUMMARY_MODEL = "Hermes-4.3-36B"
 BATCH_SIZE = int(os.getenv("BATCH_SIZE", "50"))
 SUMMARY_WINDOW_HOURS = int(os.getenv("SUMMARY_WINDOW_HOURS", "1"))
 INCLUDE_FUTURES = os.getenv("INCLUDE_FUTURES", "0").lower() in ("1", "true", "yes")
@@ -127,6 +134,20 @@ def _kv(conn, table, name) -> str:
     return (row[0] or "").strip() if row else ""
 
 
+def resolve_api_key() -> str:
+    env = os.getenv("NOUS_API_KEY", "").strip()
+    if env:
+        return env
+    try:
+        conn = psycopg2.connect(**DB_CONFIG)
+        try:
+            return _kv(conn, "api_keys", "NOUS_API_KEY")
+        finally:
+            conn.close()
+    except Exception:  # noqa: BLE001
+        return ""
+
+
 def resolve_model() -> str:
     env = os.getenv("SUMMARY_MODEL", "").strip()
     if env:
@@ -143,15 +164,15 @@ def resolve_model() -> str:
 
 
 def resolve_base_url() -> str:
-    return os.getenv("LLM_URL", DEFAULT_LLM_URL).strip() or DEFAULT_LLM_URL
+    return os.getenv("NOUS_BASE_URL", DEFAULT_NOUS_BASE_URL).strip() or DEFAULT_NOUS_BASE_URL
 
 
-def call_llm(prompt: str, *, model: str, base_url: str, json_mode: bool = False) -> str:
-    """Send a prompt to Ollama generate and return the text (\"\" on failure)."""
-    if not model:
-        logger.warning("SUMMARY_MODEL empty — skipping LLM call")
+def call_llm(prompt: str, *, api_key: str, model: str, base_url: str, json_mode: bool = False) -> str:
+    """Send a prompt to Nous chat completions and return the text (\"\" on failure)."""
+    if not api_key:
+        logger.warning("NOUS_API_KEY not set — skipping LLM call")
         return ""
-    return chat(prompt, model=model, base_url=base_url, json_mode=json_mode)
+    return chat(prompt, api_key=api_key, model=model, base_url=base_url, json_mode=json_mode)
 
 
 # ── Futures (legacy, gated) ────────────────────────────────────────────────
@@ -405,16 +426,11 @@ def summarize_news() -> None:
         )
         return
 
+    api_key = resolve_api_key()
     model = resolve_model()
     base_url = resolve_base_url()
-    if not model:
-        logger.warning("SUMMARY_MODEL empty — idle this run")
-        return
-    if not ollama_ready(model, base_url=base_url):
-        logger.warning(
-            "Ollama not ready at %s (model %s missing or /api/tags down) — idle this run",
-            base_url, model,
-        )
+    if not api_key:
+        logger.warning("NOUS_API_KEY unset in env and api_keys — idle this run")
         return
 
     articles = get_recent_news()
@@ -436,6 +452,7 @@ def summarize_news() -> None:
         )
         summary = call_llm(
             build_map_prompt(batch),
+            api_key=api_key,
             model=model,
             base_url=base_url,
             json_mode=False,
@@ -451,6 +468,7 @@ def summarize_news() -> None:
     logger.info("reduce phase over %d partial summaries", len(partial_summaries))
     master_raw = call_llm(
         build_master_prompt(final_input),
+        api_key=api_key,
         model=model,
         base_url=base_url,
         json_mode=True,
diff --git a/news/summerizer/tests/test_nous_client.py b/news/summerizer/tests/test_nous_client.py
new file mode 100644
index 0000000..002faa4
--- /dev/null
+++ b/news/summerizer/tests/test_nous_client.py
@@ -0,0 +1,97 @@
+from unittest.mock import MagicMock
+
+import httpx
+from nous_client import chat
+
+DEFAULT_UA = "osint-dashboard-news-summarizer"
+BASE = "https://inference-api.nousresearch.com/v1"
+
+
+def _ok_response(content="hello"):
+    resp = MagicMock()
+    resp.status_code = 200
+    resp.json.return_value = {"choices": [{"message": {"content": content}}]}
+    return resp
+
+
+def _install_fake(monkeypatch, post_impl):
+    captured = {}
+
+    class FakeClient:
+        def __init__(self, timeout=None, **kwargs):
+            captured["timeout"] = timeout
+
+        def __enter__(self):
+            return self
+
+        def __exit__(self, *exc):
+            return False
+
+        def post(self, url, *, headers=None, json=None, **kwargs):
+            captured["url"] = url
+            captured["headers"] = headers
+            captured["json"] = json
+            return post_impl(url, headers, json)
+
+    monkeypatch.setattr(httpx, "Client", FakeClient)
+    return captured
+
+
+def test_posts_chat_completions_with_auth_body_and_returns_content(monkeypatch):
+    captured = _install_fake(monkeypatch, lambda *a: _ok_response("the-content"))
+    out = chat(
+        "summarize this",
+        api_key="secret-key",
+        model="hermes-3",
+        base_url=BASE,
+    )
+    assert out == "the-content"
+    assert captured["url"] == f"{BASE}/chat/completions"
+    assert captured["headers"]["Authorization"] == "Bearer secret-key"
+    assert captured["headers"]["User-Agent"] == DEFAULT_UA
+    assert captured["json"]["model"] == "hermes-3"
+    assert captured["json"]["messages"] == [{"role": "user", "content": "summarize this"}]
+    assert captured["json"]["temperature"] == 0.2
+    assert captured["json"]["max_tokens"] == 4096
+    assert "response_format" not in captured["json"]
+
+
+def test_user_agent_equals_osint_user_agent_env(monkeypatch):
+    monkeypatch.setenv("OSINT_USER_AGENT", "custom-ua/2.0")
+    captured = _install_fake(monkeypatch, lambda *a: _ok_response("ok"))
+    chat("p", api_key="k", model="m", base_url=BASE)
+    assert captured["headers"]["User-Agent"] == "custom-ua/2.0"
+
+
+def test_json_mode_sets_response_format(monkeypatch):
+    captured = _install_fake(monkeypatch, lambda *a: _ok_response("{}"))
+    chat("p", api_key="k", model="m", base_url=BASE, json_mode=True)
+    assert captured["json"]["response_format"] == {"type": "json_object"}
+
+
+def test_401_returns_empty_string(monkeypatch):
+    def post_impl(*a):
+        resp = MagicMock()
+        resp.status_code = 401
+        return resp
+
+    _install_fake(monkeypatch, post_impl)
+    assert chat("p", api_key="bad", model="m", base_url=BASE) == ""
+
+
+def test_5xx_returns_empty_string(monkeypatch):
+    def post_impl(*a):
+        resp = MagicMock()
+        resp.status_code = 503
+        return resp
+
+    _install_fake(monkeypatch, post_impl)
+    assert chat("p", api_key="k", model="m", base_url=BASE) == ""
+
+
+def test_timeout_returns_empty_string(monkeypatch):
+    def post_impl(*a):
+        raise httpx.TimeoutException("timed out")
+
+    _install_fake(monkeypatch, post_impl)
+    assert chat("p", api_key="k", model="m", base_url=BASE) == ""
diff --git a/news/summerizer/tests/test_ollama_client.py b/news/summerizer/tests/test_ollama_client.py
deleted file mode 100644
index 8339c86..0000000
--- a/news/summerizer/tests/test_ollama_client.py
+++ /dev/null
@@ -1,111 +0,0 @@
-from unittest.mock import MagicMock
-
-import httpx
-from ollama_client import chat, list_tags, normalize_base, ollama_ready
-
-DEFAULT_UA = "osint-dashboard-news-summarizer"
-BASE = "http://127.0.0.1:11434"
-
-
-def _ok_response(content="hello"):
-    resp = MagicMock()
-    resp.status_code = 200
-    resp.json.return_value = {"response": content}
-    return resp
-
-
-def _install_fake(monkeypatch, post_impl=None, get_impl=None):
-    captured = {}
-
-    class FakeClient:
-        def __init__(self, timeout=None, **kwargs):
-            captured["timeout"] = timeout
-
-        def __enter__(self):
-            return self
-
-        def __exit__(self, *exc):
-            return False
-
-        def post(self, url, *, headers=None, json=None, **kwargs):
-            captured["url"] = url
-            captured["headers"] = headers
-            captured["json"] = json
-            if post_impl is None:
-                raise AssertionError("post not expected")
-            return post_impl(url, headers, json)
-
-        def get(self, url, *, headers=None, **kwargs):
-            captured["url"] = url
-            captured["headers"] = headers
-            if get_impl is None:
-                raise AssertionError("get not expected")
-            return get_impl(url, headers)
-
-    monkeypatch.setattr(httpx, "Client", FakeClient)
-    return captured
-
-
-def test_normalize_base_strips_generate_path():
-    assert normalize_base("http://ollama:11434/api/generate") == "http://ollama:11434"
-    assert normalize_base("http://ollama:11434/api/") == "http://ollama:11434"
-    assert normalize_base("http://ollama:11434") == "http://ollama:11434"
-
-
-def test_posts_generate_and_returns_response(monkeypatch):
-    captured = _install_fake(monkeypatch, post_impl=lambda *a: _ok_response("the-content"))
-    out = chat("summarize this", model="qwen3:30b-a3b", base_url=BASE)
-    assert out == "the-content"
-    assert captured["url"] == f"{BASE}/api/generate"
-    assert captured["headers"]["User-Agent"] == DEFAULT_UA
-    assert captured["json"]["model"] == "qwen3:30b-a3b"
-    assert captured["json"]["prompt"] == "summarize this"
-    assert captured["json"]["stream"] is False
-    assert "format" not in captured["json"]
-    assert captured["json"]["options"]["num_ctx"] == 32768
-    assert captured["timeout"] == 300.0
-
-
-def test_json_mode_sets_format_json(monkeypatch):
-    captured = _install_fake(monkeypatch, post_impl=lambda *a: _ok_response("{}"))
-    chat("p", model="m", base_url=BASE, json_mode=True)
-    assert captured["json"]["format"] == "json"
-
-
-def test_empty_model_returns_empty(monkeypatch):
-    assert chat("p", model="", base_url=BASE) == ""
-
-
-def test_5xx_returns_empty_string(monkeypatch):
-    def post_impl(*a):
-        resp = MagicMock()
-        resp.status_code = 503
-        return resp
-
-    _install_fake(monkeypatch, post_impl=post_impl)
-    assert chat("p", model="m", base_url=BASE) == ""
-
-
-def test_timeout_returns_empty_string(monkeypatch):
-    def post_impl(*a):
-        raise httpx.TimeoutException("timed out")
-
-    _install_fake(monkeypatch, post_impl=post_impl)
-    assert chat("p", model="m", base_url=BASE) == ""
-
-
-def test_list_tags_and_ready(monkeypatch):
-    def get_impl(*a):
-        resp = MagicMock()
-        resp.status_code = 200
-        resp.json.return_value = {
-            "models": [{"name": "qwen3:30b-a3b"}, {"name": "llama3.2:latest"}],
-        }
-        return resp
-
-    captured = _install_fake(monkeypatch, get_impl=get_impl)
-    names = list_tags(BASE)
-    assert names == ["qwen3:30b-a3b", "llama3.2:latest"]
-    assert captured["url"] == f"{BASE}/api/tags"
-    assert ollama_ready("qwen3:30b-a3b", base_url=BASE) is True
-    assert ollama_ready("missing", base_url=BASE) is False
diff --git a/tests/test_api_settings.py b/tests/test_api_settings.py
index 5d2ec40..6dad2f5 100644
--- a/tests/test_api_settings.py
+++ b/tests/test_api_settings.py
@@ -64,26 +64,31 @@ def _put(path: str, json: dict) -> httpx.Response:
     return _request("PUT", path, json=json)
 
 
-def test_get_news_models_ollama_down_returns_fallback(monkeypatch):
-    async def boom_http_get(url, *, headers, timeout):
-        raise httpx.ConnectError("ollama down")
-
-    import settings_store
-    monkeypatch.setattr(settings_store, "_http_get", boom_http_get)
-    settings_store._models_cache = None
+def test_get_news_models_without_key_returns_fallback(monkeypatch):
+    monkeypatch.delenv("NOUS_API_KEY", raising=False)
+    monkeypatch.setattr("keystore.get_api_key", _missing_key)
 
     resp = _get("/api/news/models")
     assert resp.status_code == 200
     body = resp.json()
     assert body["source"] == "fallback"
     ids = [m["id"] for m in body["models"]]
+    assert "Hermes-4.3-36B" in ids
     from settings_store import FALLBACK_MODELS
     assert ids == FALLBACK_MODELS
-    assert "qwen3:30b-a3b" in ids
 
 
-def test_get_news_models_live_from_ollama_tags(monkeypatch):
-    monkeypatch.setenv("LLM_URL", "http://ollama:11434")
+async def _missing_key(name: str):
+    return None
+
+
+async def _present_key(name: str):
+    return "test-nous-api-key-1234"
+
+
+def test_get_news_models_live_from_upstream(monkeypatch):
+    monkeypatch.setattr("keystore.get_api_key", _present_key)
+    monkeypatch.setenv("NOUS_BASE_URL", "https://inference-api.nousresearch.com/v1")
     monkeypatch.delenv("OSINT_USER_AGENT", raising=False)
 
     captured: dict = {}
@@ -95,7 +100,7 @@ def test_get_news_models_live_from_ollama_tags(monkeypatch):
             return None
 
         def json(self):
-            return {"models": [{"name": "qwen3:30b-a3b"}, {"name": "llama3.2:latest"}]}
+            return {"data": [{"id": "live-model-a"}, {"id": "Hermes-4.3-36B"}]}
 
     async def fake_http_get(url, *, headers, timeout):
         captured["url"] = url
@@ -112,10 +117,30 @@ def test_get_news_models_live_from_ollama_tags(monkeypatch):
     body = resp.json()
     assert body["source"] == "live"
     ids = [m["id"] for m in body["models"]]
-    assert ids == ["qwen3:30b-a3b", "llama3.2:latest"]
-    assert captured["url"] == "http://ollama:11434/api/tags"
+    assert ids == ["live-model-a", "Hermes-4.3-36B"]
+    assert captured["url"] == "https://inference-api.nousresearch.com/v1/models"
     assert captured["headers"]["User-Agent"] == "osint-dashboard-news-summarizer"
-    assert "Authorization" not in captured["headers"]
+    assert captured["headers"]["Authorization"] == "Bearer test-nous-api-key-1234"
+    timeout = captured["timeout"]
+    assert timeout == 8 or getattr(timeout, "read", timeout) == 8 or float(timeout) == 8.0
+
+
+def test_get_news_models_upstream_failure_returns_fallback(monkeypatch):
+    monkeypatch.setattr("keystore.get_api_key", _present_key)
+
+    async def boom_http_get(url, *, headers, timeout):
+        raise httpx.ConnectError("upstream down")
+
+    import settings_store
+    monkeypatch.setattr(settings_store, "_http_get", boom_http_get)
+    settings_store._models_cache = None
+
+    resp = _get("/api/news/models")
+    assert resp.status_code == 200
+    body = resp.json()
+    assert body["source"] == "fallback"
+    ids = [m["id"] for m in body["models"]]
+    assert "Hermes-4.3-36B" in ids
 
 
 def test_put_settings_empty_returns_422():
@@ -131,17 +156,17 @@ def test_put_settings_whitespace_only_returns_422():
 @requires_db
 def test_put_settings_round_trip(clean_settings, monkeypatch):
     monkeypatch.delenv("SUMMARY_MODEL", raising=False)
-    monkeypatch.delenv("LLM_URL", raising=False)
+    monkeypatch.delenv("NOUS_BASE_URL", raising=False)
 
-    put_resp = _put("/api/settings", {"summary_model": "qwen3:30b-a3b"})
+    put_resp = _put("/api/settings", {"summary_model": "google/gemini-2.5-flash"})
     assert put_resp.status_code == 200
     put_body = put_resp.json()
-    assert put_body["summary_model"] == "qwen3:30b-a3b"
-    assert put_body["llm_url"] == "http://127.0.0.1:11434"
+    assert put_body["summary_model"] == "google/gemini-2.5-flash"
+    assert put_body["nous_base_url"] == "https://inference-api.nousresearch.com/v1"
 
     get_resp = _get("/api/settings")
     assert get_resp.status_code == 200
     get_body = get_resp.json()
-    assert get_body["summary_model"] == "qwen3:30b-a3b"
-    assert get_body["llm_url"] == "http://127.0.0.1:11434"
-    assert set(get_body.keys()) == {"summary_model", "llm_url"}
+    assert get_body["summary_model"] == "google/gemini-2.5-flash"
+    assert get_body["nous_base_url"] == "https://inference-api.nousresearch.com/v1"
+    assert set(get_body.keys()) == {"summary_model", "nous_base_url"}