Fix synthesis fetch failures from Node's default 5-minute HTTP timeout

Nothing published for hours, every cluster failing with "fetch failed".
Ollama's own log showed the real story: requests being cancelled at
exactly 5m0s with a 500, not a model or server error. Node's global
fetch (undici) defaults to a 5-minute headers/body timeout, and CPU-only
prompt processing on the reference hardware (i5-6600K, no GPU, ~17
tok/s) legitimately takes longer than that once prompts carry full
article bodies instead of short blurbs (the previous fix in this same
line of work) — every generate() call past a few thousand tokens got
killed client-side before Ollama could finish.

OllamaProvider.generate() now passes a dedicated undici Agent with
headersTimeout/bodyTimeout disabled as the fetch dispatcher, so the
request runs as long as it actually needs to. Verified the failure
mode and the fix directly: a short-timeout dispatcher against a
deliberately slow server reproduces the exact same "fetch failed" /
UND_ERR_HEADERS_TIMEOUT error seen in production, and a zero-timeout
dispatcher completes the same slow request without issue.

undici was already a transitive dependency (via jsdom); added directly
since ollama-provider.ts now imports from it.

backend/package-lock.json

@@ -15,7 +15,8 @@
         "fastify": "^5.10.0",
         "jsdom": "^29.1.1",
         "rss-parser": "^3.13.0",
-        "telegram": "^2.26.22"
+        "telegram": "^2.26.22",
+        "undici": "^7.28.0"
       },
       "devDependencies": {
         "@types/jsdom": "^28.0.3",

backend/package.json

@@ -18,7 +18,8 @@
     "fastify": "^5.10.0",
     "jsdom": "^29.1.1",
     "rss-parser": "^3.13.0",
-    "telegram": "^2.26.22"
+    "telegram": "^2.26.22",
+    "undici": "^7.28.0"
   },
   "devDependencies": {
     "@types/jsdom": "^28.0.3",

backend/src/inference/ollama-provider.ts

@@ -1,5 +1,19 @@
+import { Agent } from 'undici';
 import type { InferenceProvider } from './provider.js';
 
+/**
+ * Node's global fetch (undici) defaults to a 5-minute headers/body timeout — fine for
+ * ordinary HTTP calls, but a real problem for /api/generate on CPU-only inference: a
+ * near-full context window can legitimately take longer than that just for prompt
+ * processing on the reference hardware (i5-6600K, no GPU, ~17 tokens/sec). Once
+ * synthesis prompts started carrying full article bodies instead of short blurbs, every
+ * generate() call past a few thousand tokens got killed at exactly 5m0s — visible in
+ * Ollama's own log as the request being cancelled, not a genuine model/server error —
+ * so no cluster could ever finish synthesizing. No timeout at all here; Ollama's own
+ * process is the natural backstop, not a clock tuned for hardware this doesn't run on.
+ */
+const noTimeoutDispatcher = new Agent({ headersTimeout: 0, bodyTimeout: 0 });
+
 /**
  * Default context window / max-generation length requested from Ollama when a caller
  * doesn't specify its own. Ollama otherwise falls back to whatever the model's
@@ -51,8 +65,11 @@ export class OllamaProvider implements InferenceProvider {
 					num_ctx: opts.numCtx ?? DEFAULT_NUM_CTX,
 					num_predict: opts.numPredict ?? DEFAULT_NUM_PREDICT
 				}
-			})
-		});
+			}),
+			// Not in the ambient RequestInit type this project resolves to, but Node's global
+			// fetch (built on undici) honors it at runtime — see noTimeoutDispatcher above.
+			dispatcher: noTimeoutDispatcher
+		} as RequestInit);
 		if (!res.ok) throw new Error(`Ollama generate failed: ${res.status} ${await res.text()}`);
 		const data = (await res.json()) as { response: string };
 		return data.response;