Translator agent

[sterile lantern]

I do not know why, but my OpenClaw do not want to translate every files i gave it. It is doing it in batches and stops, even if i asked it not to stop or wait for my input, but when i gave it something to translate everyfile i gave it (thousands of pages) it did that. It still stops and i do not know what to do.

[acoustic vault]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[sly snow]

Let me check the docs and repo structure first, then I’ll give you the cleanest answer.

I found a couple of likely limits. I’m checking the exact ones before I tell you the wrong thing.

Yeah, this sounds less like “it refuses” and more like it’s hitting limits.

Most likely causes:

• Context window overflow. Long chats + lots of files/pages cause compaction or truncation, so it starts losing track mid-job. Docs: https://docs.openclaw.ai/help/faq and https://docs.openclaw.ai/concepts/compaction
• If these are PDFs, the built-in PDF tool has real limits:
  • max 10 PDFs per call
  • 10 MB per PDF by default
  • on fallback mode, it only considers 20 pages per PDF by default
    Docs: https://docs.openclaw.ai/tools/pdf
• Old tool output can bloat the session. Context pruning helps. Docs: https://docs.openclaw.ai/concepts/session-pruning

A few important bits:

• Telling it “don’t stop” does not override model/context limits.
• If you’re using a small/local model, this gets worse fast.

What I’d check first:

1. openclaw --version
2. openclaw status --all
3. In that same chat/thread, run /status

If you paste those, I can probably tell you exactly which limit you’re hitting.

What usually fixes it:

• Start a fresh session with /new before a huge translation run
• Use /compact before continuing a long thread
• Enable context pruning
• For massive translation jobs, do smaller batches instead of “all files at once”
• If they’re PDFs, use a stronger PDF-capable model/provider

If you want, I can also help you write a sane openclaw.json config for large translation batches.

[sterile lantern]

Okay but when i asked it to translate files i gave it, even if it was thousand of pages it did everything so why not is it not working? I even created some prompts, translator agent, editor agent etc.

[sly snow]

Because “it worked once” does not mean that workflow is stable. It usually means that run happened to stay under the limits, or the model handled the chunking better that time.

The common reasons it works once and then stops later:

• The thread is now bigger. Old chat history + prior tool output eat context, so the same job can fail later in the same session. Docs: https://docs.openclaw.ai/concepts/compaction
• Different model / fallback path. If it previously used a stronger long-context or native-PDF path, and now it’s on a smaller/fallback one, behavior changes a lot.
• Different file mix. A few bigger PDFs, worse OCR, more images, or longer target-language output can push it over.
• Your translator/editor agent prompts add overhead. They help behavior, but they do not bypass context/tool limits.
• Sometimes it only looks like it translated everything, but actually skipped/summarized parts. That’s annoyingly common with giant jobs.

Also, if these are PDFs, OpenClaw’s PDF path has exact limits:

• max 10 PDFs per call
• 10 MB per PDF by default
• fallback mode only considers 20 pages per PDF by default
  Docs: https://docs.openclaw.ai/tools/pdf

So bluntly, I would not trust “thousands of pages in one chat run” as a reliable workflow.