attempt to index local metadata (a nil value)

[jakubdonovan]

I have been getting the following error very often recently when trying to send messages to the llm. Clearing chat and restarting vim doesn't solve the issue.

[magicalne]

This is due to compatibility issues with the metadata feature I developed last month. I will take a look and fix it later today.
Right now, there are three ways to resolve your issue:

1. Use tags in your config:

Plug 'magicalne/nvim.ai', { 'tag': '*' }

This will point to tag v0.1.0, which is the older version.

If you still want to stick with the main branch, then take a look at option 2 and 3.

2. Add the metadata to the head of your chat:

---
provider: deepseek
model: deepseek-chat
temperature: 0
system_prompt: You are an AI programming assistant.
---

Replace the provider and model with your own. You can overwrite these parameters directly in the chat right now.

3. Delete your cache files in the nvim.ai data folder:
   You can delete the files from the nvim plugin data folder. Since I use Linux, my data folder is located at: ~/.local/share/nvim/nvim.ai/saved_chats. You can either delete these files or move them to another folder. The chat files will then be recreated automatically.

[jakubdonovan]

This is due to compatibility issues with the metadata feature I developed last month. I will take a look and fix it later today. Right now, there are three ways to resolve your issue:

1. Use tags in your config:

Plug 'magicalne/nvim.ai', { 'tag': '*' }

This will point to tag v0.1.0, which is the older version.

If you still want to stick with the main branch, then take a look at option 2 and 3.

2. Add the metadata to the head of your chat:

---
provider: deepseek
model: deepseek-chat
temperature: 0
system_prompt: You are an AI programming assistant.
---

Replace the provider and model with your own. You can overwrite these parameters directly in the chat right now.

3. Delete your cache files in the nvim.ai data folder:
   You can delete the files from the nvim plugin data folder. Since I use Linux, my data folder is located at: ~/.local/share/nvim/nvim.ai/saved_chats. You can either delete these files or move them to another folder. The chat files will then be recreated automatically.

Thanks for getting back to me. After clearing all the relevant cache and reinstalling the plugin, I was able to open a chat but for some reason got the entire default prompt.

I have a llms-full.txt containing svelte 5 documentation, which I want the ai to learn so that It can give me higher quality answers when working with svelte projects. I know that I could simply attach the txt buffer whenever I open a new chat, but I'm thinking that it might be better to include it in the default prompt to store it in the global context. How would you approach this?

[magicalne]

You know you can overwrite system prompt.
Customizing prompt is bit of tricky. I'm thinking about put customized prompt files under nvim cache folder.
I have a trick may help a little if you don't want to update system prompt. Just insert /you and /assistant blocks before your actual questions in your projects.

{metadata}

/you
{your_long_prompt_about_svelte5}
Do you understand?

/assistant
Yes, I do.

Then you can ask questions as usual.

[magicalne]

I have a better idea. I can create a new command, such as /file or /url. This command would allow users to import a customized prompt from a file path or a URL.

[jakubdonovan]

I have a better idea. I can create a new command, such as /file or /url. This command would allow users to import a customized prompt from a file path or a URL.

Good idea! It would be nice to be able to add documentation to the prompt without overriding the prompt entirely.

[jakubdonovan]

I have a better idea. I can create a new command, such as /file or /url. This command would allow users to import a customized prompt from a file path or a URL.

I'm having a lot of issues where the ai will not respond to my messages.

Even after deleting the saved_chats folder

[magicalne]

Is there any error messages? Could you please copy/paste all your content in the chat?

[jakubdonovan]

Is there any error messages? Could you please copy/paste all your content in the chat?

No error messages - It just saves the chat. Do you want me to paste the chat here or what did you mean?

[magicalne]

Yes, just paste your chat content here. So I can try to reproduce.

[jakubdonovan]

Yes, just paste your chat content here. So I can try to reproduce.

---

provider: anthropic
model: claude-3-5-sonnet-20240620
temperature: 0
max_tokens: 4096

/system
You are an Al programming assistant.
You are currently plugged in to the Neovim text editor on a user's machine.

Your tasks include:

• Answering general programming questions.
• Explaining how the code in a Neovim buffer works.
• Reviewing the selected code in a Neovim buffer.
• Generating unit tests for the selected code.
• Proposing fixes for problems in the selected code.
• Scaffolding code for a new workspace.
• Finding relevant code to the user's query.
• Proposing fixes for test failures.
• Answering questions about Neovim.
• Running tools.

You must:

• Follow the user's requirements carefully and to the letter.
• Keep your answers short and impersonal, especially if the user responds with context outside of your tasks.
• Minimize other prose.
• Use Markdown formatting in your answers.
• Include the programming language name at the start of the Markdown code blocks.
• Avoid line numbers in code blocks.
• Avoid wrapping the whole response in triple backticks.
• Only return relevant code.

/you
/buf 21: llms-full.txt
Please carefully review the attached Svelte 5 documentation in detail. Ensure that any responses you provide regarding Svelte 5 are accurate and consistent with the information in this document. Before answering, verify your knowledge against the documentation to avoid any inaccuracies. If needed, cite specific sections or examples from the document to support your answers. Let me know if there’s anything unclear in the documentation or if you require further clarification.

/assistant:

/you:
ping

/assistant:

/you:

[magicalne]

Ok. A quick debug. Please delete content below:

/assistant:

/you:
ping

/assistant:

/you:

The first assistant response is empty. I'm not sure why that happens. But this is invalid for sure.

The last valid message of you should be:

/you
/buf 21: llms-full.txt
Please carefully review the attached Svelte 5 documentation in detail. Ensure that any responses you provide regarding Svelte 5 are accurate and consistent with the information in this document. Before answering, verify your knowledge against the documentation to avoid any inaccuracies. If needed, cite specific sections or examples from the document to support your answers. Let me know if there’s anything unclear in the documentation or if you require further clarification.

Please try to send this request again. Let's see what happens.

[jakubdonovan]

Ok. A quick debug. Please delete content below:

/assistant:

/you:
ping

/assistant:

/you:

The first assistant response is empty. I'm not sure why that happens. But this is invalid for sure.

The last valid message of you should be:

/you
/buf 21: llms-full.txt
Please carefully review the attached Svelte 5 documentation in detail. Ensure that any responses you provide regarding Svelte 5 are accurate and consistent with the information in this document. Before answering, verify your knowledge against the documentation to avoid any inaccuracies. If needed, cite specific sections or examples from the document to support your answers. Let me know if there’s anything unclear in the documentation or if you require further clarification.

Please try to send this request again. Let's see what happens.

This happens randomly. I would probably need to delete all nvim.ai cache and reinstall the plugin to fix it

[magicalne]

So sometimes it works? I cannot test anthropic right now. Are you sure it's not a network issue?
You can enable debug mode in config:

local ai = require('ai')
ai.setup({
debug=true,
...})

That will give you more information from provider.

[jakubdonovan]

So sometimes it works? I cannot test anthropic right now. Are you sure it's not a network issue? You can enable debug mode in config:

local ai = require('ai')
ai.setup({
debug=true,
...})

That will give you more information from provider.

In opts? how do I view these logs?

[magicalne]

Yes, this is correct.
There's no logs in nvim. 😢 But it will print something useful in messages. You can check messages by running cmd: :message in nvim after sending request.

[jakubdonovan]

Yes, this is correct. There's no logs in nvim. 😢 But it will print something useful in messages. You can check messages by running cmd: :message in nvim after sending request.

Apparently the prompt is too long even though I was able to send it before.

[magicalne]

Damn. I should process this situation.
Thanks for reporting!

[jakubdonovan]

Damn. I should process this situation. Thanks for reporting!

I'm not so sure about this. I was able to send this document before.

[magicalne]

I think we can estimate the size of your prompt file. Have you updated llms-full.txt? How many words in this file?

[jakubdonovan]

I think we can estimate the size of your prompt file. Have you updated llms-full.txt? How many words in this file?

I just used a much less detailed version of the docs which came up to less than half of the tokens and the prompt sent. Maybe the full version didn't send properly that time since I kept getting outdated and inaccurate answers that went directly against the contents of the docs.

[magicalne]

Alright. I just add a new feature. Now the http error message will be showed in the assistant block.

[jakubdonovan]

Alright. I just add a new feature. Now the http error message will be showed in the assistant block.

Unable to send any messages after updating

[magicalne]

Do you have two /you blocks consecutively? Looks like it's failed to parse your chat content.

[jakubdonovan]

Do you have two /you blocks consecutively? Looks like it's failed to parse your chat content.

Nope, just the one underneath the system prompt.

messages nil

[magicalne]

I cannot reproduce. 😢
This issue should be related to your chat content. messages nil means it cannot parse metadata, system prompt or messages. Can you upload your chat and llms-minimal.txt? So I can try to reproduce.

[jakubdonovan]

I cannot reproduce. 😢 This issue should be related to your chat content. messages nil means it cannot parse metadata, system prompt or messages. Can you upload your chat and llms-minimal.txt? So I can try to reproduce.

llms-minimal.txt

---

provider: anthropic
model: claude-3-5-sonnet-20240620
temperature: 0
max_tokens: 4096

/system You are an Al programming assistant. You are currently plugged in to the
Neovim text editor on a user's machine.

Your tasks include:

• Answering general programming questions.
• Explaining how the code in a Neovim buffer works.
• Reviewing the selected code in a Neovim buffer.
• Generating unit tests for the selected code.
• Proposing fixes for problems in the selected code.
• Scaffolding code for a new workspace.
• Finding relevant code to the user's query.
• Proposing fixes for test failures.
• Answering questions about Neovim.
• Running tools.

You must:

• Follow the user's requirements carefully and to the letter.
• Keep your answers short and impersonal, especially if the user responds with
  context outside of your tasks.
• Minimize other prose.
• Use Markdown formatting in your answers.
• Include the programming language name at the start of the Markdown code
  blocks.
• Avoid line numbers in code blocks.
• Avoid wrapping the whole response in triple backticks.
• Only return relevant code.

/you
/buf 29: llms-minimal.txt
Review the attached documentation in detail. Ensure
that any responses you provide regarding Svelte are accurate and consistent with
the information in this document. Before answering, verify your knowledge
against the documentation to avoid inaccuracies.

/assistant:
I understand. I have carefully reviewed the provided Svelte and SvelteKit documentation. I will ensure that any responses I give regarding Svelte and SvelteKit are accurate and consistent with this documentation. Before answering questions, I will verify my knowledge against this authoritative source to avoid providing any incorrect information. Please feel free to ask me any questions about Svelte or SvelteKit, and I will do my best to provide accurate answers based on this documentation.

/you:

[magicalne]

I cannot run with your file. But it raises another error.
I will take a close look at this issue. Meantime could you please try with a simple prompt? It's pretty late in my timezone. I will get back to you tomorrow.

[magicalne]

And make sure you are referencing the right buffer number.

/buf 29: llms-minimal.txt

[jakubdonovan]

And make sure you are referencing the right buffer number.

/buf 29: llms-minimal.txt

I'm still unable to send messages with the buffer attached. It's under half of the token limit.

After deleting it, I can send messages just fine. Why do you think this happens?

[magicalne]

are you sure 29 is the right bufnr?

[jakubdonovan]

are you sure 29 is the right bufnr?

It was when I sent it. I thought that I only needed to send it once for the ai to read it and store the information within the chat's context. Do I actually need to keep the llms-minimal.txt buffer open throughout the entire conversation?

[magicalne]

yes you need to keep it open. the command is just a placeholder. it doesn't copy the content.
the buffer num should be checked before sending requests. I will check that part later.