diff --git a/src/lib/api/api.ts b/src/lib/api/api.ts
index e74a308..73a6c62 100644
--- a/src/lib/api/api.ts
+++ b/src/lib/api/api.ts
@@ -1,6 +1,6 @@
 import { get, writable } from 'svelte/store';
 
-import { convertBlobUrlToBase64 } from '$lib/utils';
+import { convertBlobUrlToBase64, stripIncompleteSentence } from '$lib/utils';
 
 import {
     models,
@@ -223,7 +223,15 @@ export async function OL_chat(
 
         responseInProgress.set(false);
 
-        return get(pendingResponse);
+        let pending = get(pendingResponse);
+
+        if (get(chatState).values.num_predict > -1) {
+            // strip incomplete sentence from end of last message if we're restricting
+            // the number of tokens returned
+            pending.content = stripIncompleteSentence(pending.content);
+        }
+
+        return pending;
     } catch (err) {
         if (err.name !== 'AbortError') {
             console.error('OL_chat error: ', err);
diff --git a/src/lib/utils.js b/src/lib/utils.js
index 8c62680..6d8c37c 100644
--- a/src/lib/utils.js
+++ b/src/lib/utils.js
@@ -41,3 +41,25 @@ export async function convertBlobUrlToBase64(blobUrl) {
         xhr.send();
     });
 }
+
+export function stripIncompleteSentence(text) {
+    const reSentences = /[\w\n\r\*\?\.\!\"].*/gm;
+
+    let sentences = [];
+    let m;
+
+    while ((m = reSentences.exec(text)) !== null) {
+        if (m.index === reSentences.lastIndex) {
+            reSentences.lastIndex++;
+        }
+
+        m.forEach((m, gr) => {
+            console.log('m', m, 'gr', gr);
+            sentences.push(m);
+        });
+    }
+
+    sentences = sentences.filter((sent) => '!?.*"\'\n'.includes(sent.at(-1)));
+
+    return sentences.join(' ');
+}