@columenlabs/lumen-ai 0.1.266 → 0.1.268

package/dist/assistant/engine.js

@@ -100,14 +100,28 @@ function shownError(err) {
 }
 /**
  * A plain sentence for an error OpenAI sends inside the answer's stream (an `error` event, or a
- * failed response): what happened and what to do, by its code. The code is shown when it is not
- * one we know; OpenAI's own message is only logged.
+ * failed response): what happened and what to do, by its code, with OpenAI's code in brackets so
+ * the person and the logs see the real reason. A plan's usage limit and a per-minute rate limit are
+ * different things: the first waits for the plan to reset, the second only a minute. OpenAI's own
+ * message is logged (one `assistant_upstream_error` line), never shown.
  */
-function openaiStreamError(code, plan) {
-    if (/usage_limit|usage_unavailable|insufficient_quota|rate_limit/.test(code))
+function openaiStreamError(er, plan) {
+    const code = String(er.code || er.type || '');
+    console.error(JSON.stringify({
+        event: 'assistant_upstream_error',
+        provider: 'openai',
+        plan,
+        code: er.code ?? null,
+        type: er.type ?? null,
+        message: typeof er.message === 'string' ? er.message.slice(0, 500) : null,
+    }));
+    const tag = code ? ` (${code})` : '';
+    if (/usage_limit|usage_unavailable|insufficient_quota/.test(code))
         return plan
-            ? 'Your ChatGPT plan reached its usage limit for now: try again later, or pick another model.'
-            : 'OpenAI says the key is out of quota or rate limited. Try again shortly.';
+            ? `Your ChatGPT plan reached its usage limit for now${tag}: try again later, or pick another model.`
+            : `OpenAI says the key is out of quota${tag}. Add credit to the key, or pick another model.`;
+    if (/rate_limit/.test(code))
+        return `OpenAI is rate limiting these requests${tag}: try again in a minute.`;
     if (/context_length|too_long|max_tokens/.test(code))
         return 'This chat is too long for the model. Start a new chat, or type /compact.';
     if (/model_not_found|model_not_supported|invalid_model|unsupported_model/.test(code))
@@ -115,7 +129,7 @@ function openaiStreamError(code, plan) {
             ? 'This model is not on your ChatGPT plan. Pick another model for this tier in Settings.'
             : 'OpenAI does not offer this model to the key. Pick another model for this tier in Settings.';
     if (/server_error|overloaded|timeout|unavailable/.test(code))
-        return 'OpenAI is overloaded. Try again shortly.';
+        return `OpenAI is overloaded${tag}. Try again shortly.`;
     if (/content_policy|safety|refus/.test(code))
         return 'OpenAI declined to answer this message.';
     return `OpenAI could not answer (${code || 'no reason given'}). It is logged; try again or pick another model.`;
@@ -1073,7 +1087,7 @@ export class Assistant {
                     done = e.response;
                 else if (e.type === 'error') {
                     const er = (e.error ?? e);
-                    throw new AssistantError(openaiStreamError(er.code || er.type || '', plan), 'upstream');
+                    throw new AssistantError(openaiStreamError(er, plan), 'upstream');
                 }
             }
         }
@@ -1081,12 +1095,16 @@ export class Assistant {
             // The SDK throws a stream's `error` event as an APIError with no HTTP status.
             if (err instanceof OpenAI.APIError && err.status === undefined) {
                 const er = (err.error ?? {});
-                throw new AssistantError(openaiStreamError(er.code || er.type || err.code || '', plan), 'upstream');
+                throw new AssistantError(openaiStreamError({
+                    ...er,
+                    code: er.code || er.type || err.code || '',
+                    message: er.message ?? err.message,
+                }, plan), 'upstream');
             }
             throw err;
         }
         if (done?.status === 'failed')
-            throw new AssistantError(openaiStreamError(done.error?.code ?? '', plan), 'upstream');
+            throw new AssistantError(openaiStreamError(done.error ?? {}, plan), 'upstream');
         const u = done?.usage;
         if (u) {
             const read = u.input_tokens_details?.cached_tokens ?? 0;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@columenlabs/lumen-ai",
-  "version": "0.1.266",
+  "version": "0.1.268",
   "description": "Lumen's assistant and agent parts: a chat assistant's server (Claude tiers, tools over MCP, the web, hand off, spend caps), serving an MCP server from a Worker, read-only tool sets and metered tool calls.",
   "license": "MIT",
   "type": "module",
@@ -56,8 +56,8 @@
     "@ai-sdk/openai": "4.0.108",
     "@anthropic-ai/sdk": "0.133.0",
     "@cloudflare/think": "0.23.0",
-    "@columenlabs/lumen-ops": "0.1.266",
-    "@columenlabs/lumen-ui-types": "0.1.266",
+    "@columenlabs/lumen-ops": "0.1.268",
+    "@columenlabs/lumen-ui-types": "0.1.268",
     "agents": "0.32.0",
     "ai": "7.0.162",
     "just-bash": "3.6.0",