@aria-framework/ai 0.9.0 → 0.10.0

package/error.js

@@ -18,13 +18,18 @@ class AiError extends Error {
   /**
    * @param {AiErrorKind} kind
    * @param {string} message  written for the admin screen, not the log
-   * @param {{cause?: Error, status?: number}} [opts]
+   * @param {{cause?: Error, status?: number, usage?: object}} [opts]
    */
   constructor(kind, message, opts = {}) {
     super(message);
     this.name = 'AiError';
     this.kind = kind;
     this.status = opts.status || null;
+    // WHAT IT COST BEFORE IT FAILED, when the caller knows. A model that spends its whole reply
+    // budget on internal reasoning produces no answer — but it generated tokens, and something
+    // measuring throughput should not have to treat that as zero work. Optional everywhere: most
+    // failures happen before a single token is spent.
+    if (opts.usage) this.usage = opts.usage;
     if (opts.cause) this.cause = opts.cause;
   }
 

package/index.js

@@ -176,6 +176,15 @@ function createAiClient(deps = {}) {
     }
   }
 
+  /**
+   * A fixed-workload speed test for one endpoint. See benchmark.js for why the workload is fixed
+   * rather than the prompt, and why the warm-up is reported rather than discarded.
+   */
+  async function benchmarkEndpoint(cfgOverride, benchOpts) {
+    const cfg = cfgOverride || await resolveConfig();
+    return require('./benchmark').benchmark(complete, cfg, benchOpts || {});
+  }
+
   // The writing-assist engines are bound to this client's complete() so a caller gets config +
   // budget + retry for free. Prompt framing is supplied per call by the app (content).
   const boundPolish = (opts) => polish(complete, opts);
@@ -183,6 +192,7 @@ function createAiClient(deps = {}) {
 
   return {
     complete, isEnabled, test, listModels, listModelsResult, withOneRetry,
+    benchmark: benchmarkEndpoint,
     polish: boundPolish, generate: boundGenerate,
     facts, AiError, PROVIDERS, DEFAULTS
   };

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@aria-framework/ai",
   "description": "Aria App Framework — AI module. A dependency-injected model seam (createAiClient) over several providers (LM Studio / OpenAI-compatible / Anthropic), with a fact-preservation guard, generic Polish and Generate writing engines, and a browser polish widget. Prompts and config stay in the consuming app.",
-  "version": "0.9.0",
+  "version": "0.10.0",
   "license": "UNLICENSED",
   "private": false,
   "publishConfig": {
@@ -25,6 +25,6 @@
     "@aria-framework/db-worker": { "optional": true }
   },
   "scripts": {
-    "test": "node test/smoke.js && node test/usageStore.js && node test/providerStore.js && node test/health.js && node test/listModels.js && node test/views.js"
+    "test": "node test/smoke.js && node test/usageStore.js && node test/providerStore.js && node test/health.js && node test/listModels.js && node test/benchmark.js && node test/views.js"
   }
 }

package/providers/openai-compatible.js

@@ -209,11 +209,13 @@ function emptyCompletion({ label, finishReason, reasoning, maxTokens, usage }) {
     return new AiError('bad_response',
       `${label} ran out of room before it answered — it used all ${maxTokens} reply tokens` +
       (reasoning ? ' on internal reasoning' : '') +
-      '. This model thinks before it replies, so raise the reply limit (1024 is a sensible floor).');
+      '. This model thinks before it replies, so raise the reply limit (1024 is a sensible floor).',
+    { usage });
   }
   if (reasoning) {
     return new AiError('bad_response',
-      `${label} returned only its internal reasoning and no answer. Raise the reply limit and try again.`);
+      `${label} returned only its internal reasoning and no answer. Raise the reply limit and try again.`,
+    { usage });
   }
   return new AiError('bad_response',
     `${label} returned an empty completion (finish reason: ${finishReason || 'none given'}). ` +

package/views/ai/provider-card.ejs

@@ -79,9 +79,23 @@
             <span class="font-monospace"><%= Number(usage.total_tokens || 0).toLocaleString() %></span></div>
         <% } %>
         <div>
-          <%# LAST GOOD, not just a dot. "last good 3h ago" tells a story a green light cannot. %>
+          <%# LAST GOOD, not just a dot. "last good 3h ago" tells a story a green light cannot.
+              THREE STATES, NOT TWO. `ago(...) || 'never'` claimed an endpoint had never worked
+              whenever there was no timestamp — but health lives in process memory, so every
+              restart wiped the record and a perfectly healthy endpoint came back reading "never".
+              That is not a missing value, it is a DIFFERENT FACT, and stating the stronger one is
+              how an operator ends up debugging a server that was fine. The checker already tells
+              them apart: `status: 'unknown'` means it has not been seen in this process at all,
+              whereas a real entry with no lastGoodAt has genuinely been tried and never answered. %>
           <div class="text-body-secondary text-uppercase" style="font-size:.68rem;letter-spacing:.06em">Last good</div>
-          <span class="font-monospace"><%= ago(h.lastGoodAt) || 'never' %></span>
+          <% if (h.lastGoodAt) { %>
+            <span class="font-monospace"><%= ago(h.lastGoodAt) %></span>
+          <% } else if (h.status === 'unknown') { %>
+            <span class="font-monospace text-body-secondary"
+                  title="Nothing recorded since this server started. Health is kept in memory, so a restart clears it — press Test to find out.">no record</span>
+          <% } else { %>
+            <span class="font-monospace">never</span>
+          <% } %>
         </div>
       </div>