diff --git a/src/ai/ai_client.c b/src/ai/ai_client.c
index 00f8d007..1187ebc1 100644
--- a/src/ai/ai_client.c
+++ b/src/ai/ai_client.c
@@ -35,6 +35,9 @@ static Autocomplete providers_ac = NULL;
 static AIProvider* _ai_add_provider_nopersist(const gchar* name, const gchar* api_url);
 static void _ai_load_models_for_provider(AIProvider* provider);
 static void _ai_save_models_for_provider(AIProvider* provider);
+static const gchar* _find_json_field(const gchar* json, const gchar* field);
+static gchar* _extract_json_string(const gchar* json, const gchar* field);
+static gchar* _parse_error_response(const gchar* error_json);
 
 /* ========================================================================
  * Curl helpers
@@ -184,6 +187,59 @@ ai_json_escape(const gchar* str)
     return g_string_free(result, FALSE);
 }
 
+/* Decode JSON string escapes in [start, end) into a freshly allocated buffer.
+ * Handles \" \\ \/ \n \t \r \b \f. \uXXXX sequences are passed through verbatim
+ * since the rest of the AI pipeline already deals in UTF-8 byte streams. */
+static gchar*
+_json_unescape_substring(const gchar* start, const gchar* end)
+{
+    gchar* out = g_new0(gchar, (end - start) + 1);
+    gchar* w = out;
+    for (const gchar* in = start; in < end;) {
+        if (*in == '\\' && in + 1 < end) {
+            switch (in[1]) {
+            case '"':  *w++ = '"';  in += 2; break;
+            case '\\': *w++ = '\\'; in += 2; break;
+            case '/':  *w++ = '/';  in += 2; break;
+            case 'n':  *w++ = '\n'; in += 2; break;
+            case 't':  *w++ = '\t'; in += 2; break;
+            case 'r':  *w++ = '\r'; in += 2; break;
+            case 'b':  *w++ = '\b'; in += 2; break;
+            case 'f':  *w++ = '\f'; in += 2; break;
+            default:   *w++ = *in++; break;
+            }
+        } else {
+            *w++ = *in++;
+        }
+    }
+    *w = '\0';
+    return out;
+}
+
+/* Locate "field":"...value..." in json and return the unescaped value.
+ * Skips JSON whitespace around the colon and treats any \X (X != \0) inside
+ * the string body as an opaque escape so an embedded \" cannot be misread as
+ * the closing quote. */
+static gchar*
+_extract_json_string(const gchar* json, const gchar* field)
+{
+    const gchar* val = _find_json_field(json, field);
+    if (!val || *val != '"')
+        return NULL;
+
+    const gchar* start = val + 1;
+    for (const gchar* p = start; *p; p++) {
+        if (*p == '\\' && *(p + 1) != '\0') {
+            p++; /* skip the escaped character; loop ++ advances past it */
+            continue;
+        }
+        if (*p == '"') {
+            return _json_unescape_substring(start, p);
+        }
+    }
+    return NULL;
+}
+
 /* ========================================================================
  * Provider Management
  * ======================================================================== */
@@ -616,8 +672,11 @@ _curl_exec_and_handle(CURL* curl, const gchar* url, struct curl_slist* headers,
         log_error("Request failed for '%s': %s", provider->name, error_msg);
         _aiwin_display_error(user_data, error_msg);
     } else if (http_code >= 400) {
-        auto_gchar gchar* error_msg = g_strdup_printf("HTTP %ld: %s", http_code,
-                                                      response->data ? response->data : "Unknown error");
+        auto_gchar gchar* parsed = _parse_error_response(response->data);
+        auto_gchar gchar* error_msg = parsed
+            ? g_strdup_printf("HTTP %ld: %s", http_code, parsed)
+            : g_strdup_printf("HTTP %ld: %s", http_code,
+                              response->data && strlen(response->data) > 0 ? response->data : "Unknown error");
         log_error("Request HTTP error for '%s': %s", provider->name, error_msg);
         _aiwin_display_error(user_data, error_msg);
     } else {
@@ -1272,139 +1331,30 @@ _parse_ai_response(const gchar* response_json)
     if (!response_json || strlen(response_json) == 0)
         return NULL;
 
-    /* Try Perplexity /v1/agent format first: look for "text":"..." inside output array
-     * Response: {"output":[{"content":[{"text":"...","type":"output_text"}]}]} */
-    const gchar* text_start = strstr(response_json, "\"text\":\"");
-    if (text_start) {
-        text_start += strlen("\"text\":\"");
-        const gchar* p = text_start;
-        while (*p) {
-            if (*p == '\\' && *(p + 1) == '"') {
-                p += 2;
-                continue;
-            }
-            if (*p == '"') {
-                gsize len = p - text_start;
-                gchar* result = g_new0(gchar, len + 1);
-                gchar* out = result;
-                const gchar* in = text_start;
-                while (in < p) {
-                    if (*in == '\\' && *(in + 1) == '"') {
-                        *out++ = '"';
-                        in += 2;
-                    } else if (*in == '\\' && *(in + 1) == 'n') {
-                        *out++ = '\n';
-                        in += 2;
-                    } else {
-                        *out++ = *in++;
-                    }
-                }
-                *out = '\0';
-                return result;
-            }
-            p++;
-        }
-    }
+    /* Perplexity /v1/agent: {"output":[{"content":[{"text":"...","type":"output_text"}]}]}
+     * Legacy OpenAI:        {"choices":[{"message":{"content":"..."}}]} */
+    gchar* text = _extract_json_string(response_json, "text");
+    if (text)
+        return text;
 
-    /* Try legacy OpenAI format: "content":"..."
-     * Response: {"choices":[{"message":{"content":"..."}}]} */
-    const gchar* content_start = strstr(response_json, "\"content\":\"");
-    if (!content_start)
-        return NULL;
-
-    content_start += strlen("\"content\":\"");
-
-    /* Find the closing quote, accounting for escaped quotes */
-    const gchar* p = content_start;
-    while (*p) {
-        if (*p == '\\' && *(p + 1) == '"') {
-            /* Escaped quote, skip both characters */
-            p += 2;
-            continue;
-        }
-        if (*p == '"') {
-            /* Found unescaped closing quote */
-            gsize len = p - content_start;
-            /* Unescape the content: convert \" back to " */
-            gchar* result = g_new0(gchar, len + 1);
-            gchar* out = result;
-            const gchar* in = content_start;
-            while (in < p) {
-                if (*in == '\\' && *(in + 1) == '"') {
-                    *out++ = '"';
-                    in += 2;
-                } else if (*in == '\\' && *(in + 1) == 'n') {
-                    *out++ = '\n';
-                    in += 2;
-                } else {
-                    *out++ = *in++;
-                }
-            }
-            *out = '\0';
-            return result;
-        }
-        p++;
-    }
-
-    return NULL;
+    return _extract_json_string(response_json, "content");
 }
 
-/**
- * Extract error message from an API error JSON response.
- * Handles format: {"error":{"message":"...","type":"...","code":...}}
- */
+/* Extract the human-readable error.message from a provider error envelope:
+ *   {"error":{"message":"...","type":"...","code":...}}
+ * Scoped to the substring after "error": so unrelated top-level "message"
+ * fields do not get picked up. */
 static gchar*
 _parse_error_response(const gchar* error_json)
 {
     if (!error_json || strlen(error_json) == 0)
         return NULL;
 
-    /* Look for "message":"..." inside "error" object */
     const gchar* error_obj = strstr(error_json, "\"error\":");
     if (!error_obj)
         return NULL;
 
-    const gchar* message_key = strstr(error_obj, "\"message\":\"");
-    if (!message_key)
-        return NULL;
-
-    message_key += strlen("\"message\":\"");
-    const gchar* msg_start = message_key;
-    const gchar* p = msg_start;
-    while (*p) {
-        if (*p == '\\' && *(p + 1) == '"') {
-            p += 2;
-            continue;
-        }
-        if (*p == '"') {
-            gsize len = p - msg_start;
-            gchar* result = g_new0(gchar, len + 1);
-            gchar* out = result;
-            const gchar* in = msg_start;
-            while (in < p) {
-                if (*in == '\\' && *(in + 1) == '"') {
-                    *out++ = '"';
-                    in += 2;
-                } else if (*in == '\\' && *(in + 1) == 'n') {
-                    *out++ = '\n';
-                    in += 2;
-                } else if (*in == '\\' && *(in + 1) == '\\') {
-                    *out++ = '\\';
-                    in += 2;
-                } else if (*in == '\\' && *(in + 1) == 't') {
-                    *out++ = '\t';
-                    in += 2;
-                } else {
-                    *out++ = *in++;
-                }
-            }
-            *out = '\0';
-            return result;
-        }
-        p++;
-    }
-
-    return NULL;
+    return _extract_json_string(error_obj, "message");
 }
 
 static gpointer