Troubleshooting Errors in Your Output

If your Assist 2.0 output is throwing errors, slowing down, or returning a "something went wrong" message, you've likely hit one of two common issues: too many images dropped into a single chat, or a conversation that's filled up its memory. Both are easy to fix once you know what to look for.

This article walks through:

• Why outputs return errors
• How to upload heavy content the right way (via Knowledge)
• How to track your conversation's memory usage
• What to do when memory fills up — starting a fresh chat without losing context

1. Why your output is returning errors

When an output errors out, you'll usually see a message along the lines of:

"Sorry, something went wrong. Let me know if you want me to try again, or you can contact your account manager."

Note: your organisation may have customised this message via always-include org knowledge, so the exact wording can vary.

There are two main causes:

• Too many images uploaded directly into the chat — the underlying vision model has hard per-request limits. Exceeding any of them causes the output to fail.
• The conversation's memory is full or close to full — as memory approaches 100%, responses slow down and errors become more likely.

Image limits (per Anthropic's vision specs)

💡 Pro tip: Assist 2.0's vision works effectively at ~1080p, so downscaling large images rarely affects analysis quality and saves memory.

2. The fix — Upload heavy content to Knowledge

Instead of attaching images or large files to a chat message, add them to Knowledge. Assist 2.0 can read them from there without consuming chat memory, and they'll be available across every output in the workspace.

Step-by-step

1. From the left navigation, open Library.
2. In the Library sub-nav, click the Knowledge tab.
3. Click + Add Knowledge (or the equivalent create button).
4. Choose your source — options include uploading a file, pasting text, scanning a URL, or uploading an image.
5. Select your files and confirm the upload. You'll see a processing indicator with thumbnails and a progress bar.
6. Give the knowledge a clear Title (e.g. Campaign reference images).
7. Save. You should see a confirmation that knowledge has been saved successfully.

Each uploaded item will appear in the Knowledge list with a Training in Progress status, which changes to Training Complete once Assist 2.0 has indexed it.

[SCREENSHOT: Library → Knowledge tab with the + Add Knowledge flow open]

⚠️ Heads-up on workspace scoping: Knowledge is scoped to the workspace you're working in. If you can't see a piece of knowledge you expected, double-check you're in the right workspace.

3. Tracking conversation memory

Every Assist 2.0 conversation has a memory limit (also called the context window). The full context window is 400K tokens.

How to check your memory usage

1. Open any output or chat.
2. Look next to the send button in the message input — you'll see a percentage (e.g. 0%  ).
3. Hover over the percentage to see the full tooltip, which displays both a percentage and a token count (e.g. 0% — 35 / 400K  ).

[SCREENSHOT: Message input showing the memory % indicator and tooltip]

What to watch for

• Below ~80% — Conversation runs normally.
• Approaching 100% — Responses may slow down and errors become more likely.
• At 100% — Assist 2.0 automatically compacts the conversation, compressing older messages to free up space.

Compaction preserves the gist of the conversation but can reduce the precision of older context. For long, complex tasks, it's better to start fresh before compaction kicks in.

4. When memory fills up — Start a fresh chat

If your conversation is approaching the memory limit or has already been compacted, the cleanest fix is to start a new output. In Assist 2.0, you don't need to manually carry context across — the platform searches existing projects, knowledge, templates, and agents in your workspace when planning a response, so prior work remains accessible.

How to continue your work cleanly

1. From your project, click + New Output (or start a new chat via Just Start Talking).
2. In the chat, describe what you want to continue working on in natural language. Reference the previous output or topic by name — Assist 2.0 will locate relevant prior projects, outputs, and knowledge automatically.
3. If there's a specific file or output you want pulled in directly, use the + icon next to the message input to attach a file, paste a URL, or upload platform data.

The new chat starts with 0%   memory used, and Assist 2.0 will pull in the relevant context from your workspace as needed.

💡 Tip: For anything you'll want to reuse across multiple chats — reference docs, brand guidelines, image libraries — put it in Knowledge rather than relying on chat history. That way it's always available, and it doesn't eat into your memory budget.

Best practices summary

✅ Do

• Upload images and heavy files to Knowledge, not directly to chat
• Keep image batches under 20 per message and 100 per request
• Downscale large images to ~1080p before uploading
• Monitor the memory % indicator regularly on long conversations
• Start a fresh chat when memory gets heavy — Assist 2.0 will reuse relevant workspace assets

❌ Don't

• Drag dozens of images into a chat message
• Ignore the memory tracker until you hit 100%
• Keep working in a chat that's already showing errors — start fresh instead

Related articles

• Creating and Managing Knowledge in Assist 2.0
• Just Start Talking: the unified chat interface
• Understanding context windows in Assist 2.0

Still need help?

If you continue to see errors after following these steps, contact your account manager.