V0.7.6 first start setup | CONFIG_KEY_NOT_FOUND

[slu-010101]

Hi Keith,

I saw on two different users that WriterAgent don't work after first setup:

2026-04-30 16:02:06,924 | plugin.modules.http.requests | ERROR | HTTP Error: HTTP-Fehler 401 vom KI-Anbieter: Unauthorized.
Provider Response:
unauthorized: invalid or missing API key
2026-04-30 16:02:06,924 | plugin.framework.config | WARNING | fetch_available_models NetworkError for https://localai/v1/models: HTTP-Fehler 401 vom KI-Anbieter: Unauthorized.

File "/home/userA/.config/libreoffice/4/user/uno_packages/cache/uno_packages/lu54024hwfln.tmp_/WriterAgent v0.7.6.oxt/plugin/framework/config.py", line 418, in _resolve_default
    raise ConfigError(f"Config key {key!r} not found in schema", "CONFIG_KEY_NOT_FOUND")

Closing LibreOffice after setup and start it again fix this issue. Is it possible there is a bug in the first configuration process since I enter first the localai server and in the next step the API key?

The new button to open the JSON config is awesome!

Sam

[KeithCu]

Hi Sam, thank you for the input!! I think I fixed the 1st issue. As for the second, I see you do use security for local AI, I'll investigate that issue. Okay I'm making a new build of 0.7.6, if you want to see if it fixes the issues, that would be great, let me know!

[slu-010101]

Hi Keith,

yes this problem is fixed and also the models are now automatically loaded (doesn't working before) as soon I input my API key.

But unfortunately it doesn't format my document any more:

2026-05-01 17:43:55,040 | plugin.modules.chatbot.librarian | ERROR | Librarian error: Error while generating output:
HTTP-Fehler 500 vom KI-Anbieter: Internal Server Error. Failed to parse input at pos 274: <|channel|>final <|constrain|>commentary<|message|>Action:

Do you have any idea what the cause might be?

Thanks.

Sam

[KeithCu]

@slu-010101 Hi Sam, Thank you for reporting that the issue is fixed and sorry to hear you have another bug. I will look into it before the next release. Was it a new install? (The librarian only runs once.) There is a parsing error. I'm not sure what changed that would cause it to happen but I did make some small fixes and improved the error logging. Can you try the new build when you have time?

Thank you so much for your help!!

[KeithCu]

BTW, you were exactly correct that I didn't re-run the model fetch after the user put in the api key, but as you noticed I changed it now to handle that case. If you have any more issues or errors, please let me know! I don't have good hardware for doing AI so when I test via OpenRouter or Together.ai, it all works perfectly, but there is no harmony tool call parsing,etc. I want to make sure you are happy before I make my next release!

[slu-010101]

Not working, unfortunately:

2026-05-01 23:20:30,884 | writeragent.grammar | INFO | [grammar] ensure_proofreader_selection: entry
2026-05-01 23:20:30,884 | writeragent.grammar | INFO | [grammar] ensure_proofreader_selection: Doc-tab AI grammar off (enable on Doc tab to use the checker)
2026-05-01 23:20:30,998 | plugin.framework.document | WARNING | get_document_property error: Get property value fallback failed: WriterAgentSessionID
2026-05-01 23:22:07,208 | plugin.modules.http.client | ERROR | Provider API Error 500: {"error":{"code":500,"message":"Failed to parse input at pos 206: <|channel|>final <|constrain|>commentary<|message|>{\n  \"name\": \"reply_to_user\",\n  \"arguments\": {\n    \"answer\": \"Viel Erfolg beim Weiterarbeiten, Sam! 😊\"\n  }\n}","type":"server_error"}}
2026-05-01 23:22:07,214 | plugin.modules.chatbot.librarian | ERROR | Librarian error: Error while generating output:
HTTP-Fehler 500 vom KI-Anbieter: Internal Server Error. Failed to parse input at pos 206: <|channel|>final <|constrain|>commentary<|message|>{
  "name": "reply_to_user",
  "arguments": {
    "answer": "Viel Erfolg beim Weiterarbeiten, Sam! 😊"
  }
}

I tried it always with a new clean LibreOffice profile and the last version (check the sha256 sum).

[KeithCu]

The log files are so helpful. I will investigate it more this weekend and make a fix. Thank you for letting me know!!

[KeithCu]

Hi @slu-010101 I made a new build, can you try it out? What's weird is that I added new code so I'm not sure how it could have worked in the past? d1dfdc4

I also added more logging around the failure cases also just in case to help.

Can you try it out when you have time and let me know how it goes?

[slu-010101]

Thank you @KeithCu last version working again.
During testing it tells me no compatible document found in the window, any idea what's wrong?

Note: The list of the states was generated by the WriterAgent before.

[KeithCu]

@slu-010101 Hi Sam, Thank you for the good news and the bug report. I can't repro the issue but I see a change that could possibly have caused the issue - but it was many ago! I made a new build that should fix it if that issues happens: b5f7df0 and also it has very detailed logging for that error also.

I made a new build, can you please try it out when you have time? , I'm sorry you run into issues that I don't see. I've got 900 test cases, plus I use the app daily, so it's been stable for me. The core code is not changing much now (mostly the specific modules like grammar checker, specific handler for math, or other UNO, etc.) so there shouldn't be too many regression happening.

Thanks again for the help! Have you tried the grammar checker yet? It's not perfect yet, but it should be pretty stable and efficient and smart. Also, have you tried out the math? AI's are awesome at TeX, so the code works well now:

[slu-010101]

I made a new build, can you please try it out when you have time?

@KeithCu yes the new build fix this issue, thank you very much for this super fast bug fix!

I'm sorry you run into issues that I don't see.

No worries please, I'm very happy with the add-on and must thank you for the work on open source software.
Unfortunately I can't code, but if there is any bug, early or later I will find it... :)

Have you tried the grammar checker yet?

Not yet, it's on my to do list for next week.

Thank you for this add-on.

[KeithCu]

Okay very good, and I'm glad to hear you are happy with the add-on. With your approval I will make a new release for the LO website and GH.

Good you didn't test the grammar checker yet, it took some time to get the architecture correct since it's async but the LO method is sync, and that required a bit of work. (I return no errors, and then look it up in the background, and then when the inevitably ask again as you keep typing, then I report the errors. I've not done a lot of testing, but it works!!) There's a chance some errors miss some times, or take a little while to show up, but those are not huge problems. It's pretty stable now for 34 languages so turn it on when you have time to play with it! I tested it with English, Spanish and Japanese ;-) the runtime behavior section of the document describes what features I needed to make it work well between the sync versus async design. It needs more work as also discussed in the document but it's useful now.

https://github.com/KeithCu/writeragent/blob/master/docs/realtime-grammar-checker-plan.md

Also be sure to try this at some point: "Please review my document and add comments." Have you seen that? You might not need to prompt it to add comments, it should just trigger on "Review, give feedback or suggestions", but specific tool references always make the advice stronger "Do web research and write a colorful, detailed report on...". Try the review feature. I told LLM to put in both positive and negative comments, to make it more fun.

The new release is up there, check it out now.

BTW, I wrote 3 articles about WriterAgent, this is the first, you'll find the links to the others in there: https://keithcu.com/wordpress/?p=5060

Go through it if you are interested because I try to make them educational and to both programmers and regular people. I referenced you a couple of times ;-) I'm working on a new post for the grammar checker since it's a challenging problem.

BTW I never make a new user config directory anymore, so don't bother. If you see a bug, it will be in a new or fresh config. If you just take the time to give me the error message and relevant log entries, that should tell me enough, and if not I'll make a new build with better logging.

I have a ton of debug + error logging right now, but there are edge cases all over the project, and in fact this situation of harmony errors showing up isn't even a real error on my side, except by the LLM!

It was giving us extra special prompt template strings at the end of the tool call requests, basically garbage, and we were sending them back to them in the next message as you do in a conversation, figuring it was valid, and it wasn't liking them because they are special strings, so it was complaining. So I just strip them out.

Thank you, and enjoy the new version and let me know how it goes.

[slu-010101]

@KeithCu unfortunately there are more issues, sorry!

2026-05-04 15:46:15,897 | plugin.modules.chatbot.librarian | ERROR | Librarian error: Error while generating output:
HTTP-Fehler 500 vom KI-Anbieter: Internal Server Error. Failed to parse input at pos 156: <|channel|>final <|constrain|>json<|message|>{
  "name": "reply_to_user",
  "arguments": {
    "answer": "

    2026-05-04 15:48:18,980 | plugin.modules.chatbot.librarian | ERROR | Librarian error: Error while generating output:
HTTP-Fehler 500 vom KI-Anbieter: Internal Server Error. Failed to parse input at pos 274: <|channel|>final <|constrain|>answer<|message|>{
  "name": "reply_to_user",
  "arguments": {
    "answer":

[KeithCu]

Ah, I see. I will research it when I have time. Thank you for the report! BTW. for now, just dig up an old USER.md and it will skip past the librarian ;-)

Update: I did research, it's another case of those leaked template tokens! They send them to us, but if we send the back in the conversation history, they get mad! I'll make another fix.

[KeithCu]

@slu-010101 Okay try the new release on GitHub when you have a chance. Thank you for the report!!

[slu-010101]

@KeithCu unfortunately it doesn't working, I tried it with an empty/new LibreOffice profile.

2026-05-05 09:34:32,609 | plugin.modules.chatbot.librarian | ERROR | Librarian error: Error while generating output:
HTTP-Fehler 500 vom KI-Anbieter: Internal Server Error. Failed to parse input at pos 140: <|channel|>final <|constrain|>commentary<|message|>Action:
{
  "name": "reply_to_user",
  "arguments": {
    "answer": "Hallo! 👋 Ich bin der WriterAgent Librarian – Ihr freundlicher Begleiter, um mehr über Sie zu erfahren und Ihnen zu zeigen, was unser WriterAgent alles kann. Wie darf ich Sie nennen?"
  }
}

[KeithCu]

Hi Sam, thank you for the report. I did more research and I can't see any issue on my side. It's those same harmony tokens but I strip them and never send them so I don't know how it can complain!!

Can you please try this debug build when you have time which gives me more info? Delete the log file before you try the repro so it's small.

This is a ZIP, I can't upload OXT in chat. So just rename first ;-)

WriterAgent.zip

Let me know what you find when you have time. Thank you!!

[slu-010101]

Here are the output of WriterAgent:

KI: Ich kann Ihr Dokument sofort mit professioneller Formatierung und Farbe bearbeiten oder übersetzen. Probieren Sie es aus!

You: Hi!
[Thinking] Step 1:

 /thinking

[Bibliothekar-Fehler: Librarian failed: Error while generating output:
HTTP-Fehler 500 vom KI-Anbieter: Internal Server Error. Failed to parse input at pos 140: <|channel|>final <|constrain|>commentary<|message|>Action:
{
  "name": "reply_to_user",
  "arguments": {
    "answer": "Hallo! 👋 Ich bin dein freundlicher WriterAgent Librarian. Wie darf ich dich nennen?"
  }
}

Traceback (most recent call last):
  File "/home/sam/.config/libreoffice/4/user/uno_packages/cache/uno_packages/lu121918m4now.tmp_/WriterAgent_DEBUG.oxt/plugin/contrib/smolagents/agents.py", line 979, in _step_stream
    chat_message: ChatMessage = self.model.generate(
                                ~~~~~~~~~~~~~~~~~~~^
        input_messages,
        ^^^^^^^^^^^^^^^
        stop_sequences=["Observation:", "Calling tools:"],
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
        tools_to_call_from=self.tools_and_managed_agents,
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "/home/sam/.config/libreoffice/4/user/uno_packages/cache/uno_packages/lu121918m4now.tmp_/WriterAgent_DEBUG.oxt/plugin/framework/smol_model.py", line 52, in generate
    result = self.api.request_with_tools(
        msg_dicts,
    ...<4 lines>...
        prepend_dev_build_system_prefix=False,
    )
  File "/home/sam/.config/libreoffice/4/user/uno_packages/cache/uno_packages/lu121918m4now.tmp_/WriterAgent_DEBUG.oxt/plugin/modules/http/client.py", line 1031, in request_with_tools
    raise NetworkError(
    ...<3 lines>...
    )
plugin.framework.base_errors.NetworkError: HTTP-Fehler 500 vom KI-Anbieter: Internal Server Error. Failed to parse input at pos 140: <|channel|>final <|constrain|>commentary<|message|>Action:
{
  "name": "reply_to_user",
  "arguments": {
    "answer": "Hallo! 👋 Ich bin dein freundlicher WriterAgent Librarian. Wie darf ich dich nennen?"
  }
}

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/home/sam/.config/libreoffice/4/user/uno_packages/cache/uno_packages/lu121918m4now.tmp_/WriterAgent_DEBUG.oxt/plugin/modules/chatbot/librarian.py", line 154, in execute
    for step in run_stream:
                ^^^^^^^^^^
  File "/home/sam/.config/libreoffice/4/user/uno_packages/cache/uno_packages/lu121918m4now.tmp_/WriterAgent_DEBUG.oxt/plugin/contrib/smolagents/agents.py", line 618, in _run_stream
    raise e
  File "/home/sam/.config/libreoffice/4/user/uno_packages/cache/uno_packages/lu121918m4now.tmp_/WriterAgent_DEBUG.oxt/plugin/contrib/smolagents/agents.py", line 600, in _run_stream
    for output in self._step_stream(action_step):
                  ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
  File "/home/sam/.config/libreoffice/4/user/uno_packages/cache/uno_packages/lu121918m4now.tmp_/WriterAgent_DEBUG.oxt/plugin/contrib/smolagents/agents.py", line 995, in _step_stream
    raise AgentGenerationError(f"Error while generating output:\n{e}", self.logger) from e
plugin.contrib.smolagents.utils.AgentGenerationError: Error while generating output:
HTTP-Fehler 500 vom KI-Anbieter: Internal Server Error. Failed to parse input at pos 140: <|channel|>final <|constrain|>commentary<|message|>Action:
{
  "name": "reply_to_user",
  "arguments": {
    "answer": "Hallo! 👋 Ich bin dein freundlicher WriterAgent Librarian. Wie darf ich dich nennen?"
  }
}

And in the debug log:

2026-05-05 18:19:27,038 | plugin.modules.http.client | DEBUG | Opening new connection to https://localai:443
2026-05-05 18:19:39,738 | plugin.modules.http.client | ERROR | Provider API Error 500: {"error":{"code":500,"message":"Failed to parse input at pos 140: <|channel|>final <|constrain|>commentary<|message|>Action:
[...]

Is the "/v1" missing here?
Does that help you?

Edit:
Server logs:

192.168.77.20 - - [05/May/2026:18:19:15 +0200] "GET /v1/models HTTP/1.1" 200 544 "-" "WriterAgent (https://github.com/KeithCu/writeragent)"
192.168.77.20 - - [05/May/2026:18:19:39 +0200] "POST /v1/chat/completions HTTP/1.1" 500 320 "-" "-"

[KeithCu]

Those logs were very helpful. Please try this new debug build!
WriterAgent.zip

[slu-010101]

2026-05-05 19:43:59,626 | plugin.modules.http.client | DEBUG | Closing persistent connection to ('https', 'localai', 443, 'unverified')
2026-05-05 19:43:59,634 | plugin.modules.chatbot.librarian | ERROR | Librarian error: Error while generating output:
HTTP-Fehler 500 vom KI-Anbieter: Internal Server Error. Failed to parse input at pos 614: <|channel|>final <|constrain|>answer<|message|>{
  "name": "reply_to_user",
  "arguments": {
    "answer": "Hier ist die Liste aller deutschen Bundesländer, die du in dein Dokument einfügen kannst:\n\n1. Baden‑Württemberg\n2. Bayern\n3. Berlin\n4. Brandenburg\n5. Bremen\n6. Hamburg\n7. Hessen\n8. Mecklenburg‑Vorpommern\n9. Niedersachsen\n10. Nordrhein‑Westfalen\n11. Rheinland‑Pfalz\n12. Saarland\n13. Sachsen\n14. Sachsen‑Anhalt\n15. Schleswig‑Holstein\n16. Thüringen\n\nDu kannst diese Aufzählung einfach kopieren und an der gewünschten Stelle in dein Dokument einfügen. Viel Erfolg! 🚀"
  }
}

I use this .zip you posted (ebb17edbcc3dec4ac32634f8af1974668ebaac382bdebe10601989411dadb82d).

[KeithCu]

Can you please paste more lines above the 500 error? I want to see full request and params.

BTW, the /v1 is not an issue, it's more with the request shape, little details that have changed in the parameters. This is an interesting problem, I appreciate your help tracking this down!

[slu-010101]

Here are more logs above the 500 error:

writeragent_debug.txt

[KeithCu]

Those logs are so helpful! I researched it and found a couple of issues. There was a little bug where I didn't stop the smolagents loop immediately after switch_to_document_mode succeeds, preventing the unnecessary extra LLM call that was failing on LocalAI.

I so appreciate your help. Please try this out. It's another debug build.

Please let me know what you find!
WriterAgent.zip

Remember, you can delete the USER.md to trigger the librarian on the next startup, no need to wipe the LO config.

TY!!

[slu-010101]

Yes it's working now, tested with a new LibreOffice profile.
Thank you!

[KeithCu]

Great news!!

I put in a test case for this issue and every other bug to keep things always more stable. Usually the error logs are enough but this one required the DEBUG level to see more info on what happened before the error.

I'm glad you are happy with the program. Let me know how it goes with Add comment feature and the grammar checker. I'm very curious to hear. I love both of those features.

[KeithCu]

BTW, I just added a new feature: Enter to Send / Record. Shift+enter for CR. There's an option for now to turn it off, but I think most people will like it and usually don't put CRs in there. It's on GH now if you want to try.

[KeithCu]

Okay this is all fixed, I'll close this for now. Create new discussions with any new issues. I'm making many changes to the grammar but it should be a good prototype what you have. I was using sqlite but I'm now persisting in the document (with the version on GH) instead of sqlite. There are pros and cons to both approaches but I like keeping it with the document so that if you open files on separate computers, they will all share the errors!