Invalid schema for function 'vectorcode'

[ahakanbaba]

Thank you very much for your time in advance.

I am setting up vectorcode for the first time, so apologies is this is a trivial configuration mistake. Followed the instructions here

When I try to use @vectorcode in the Chat window, I get this error printed to neovim messages

[ERROR] 2025-05-11 04:56:11
Error: {
  "error": {
    "message": "Invalid schema for function 'vectorcode': In context=('properties', 'options'), 'required' is required to be supplied and to be an array including every key in properties. Missing 'project_root'.",
    "type": "invalid_request_error",
    "param": "tools[0].function.parameters",
    "code": "invalid_function_parameters"
  }
}

I should be using the latest versions of codecompanion and vectorcode

cat ./nvim/lazy-lock.json | grep -Ei "vector|codecom"
  "VectorCode": { "branch": "main", "commit": "2815496b98d9002c83d721c71412062196d73b85" },
  "codecompanion.nvim": { "branch": "main", "commit": "61753270f06ace9d0615f343c898ade7468eac2d" },

From a terminal i can successfully run

vectorcode ls and vectorcode query "something" . Namely the commands from the CLI are working.

I looked around a little bit and I could not be sure whether the issue should be in codecompanion or vectorcode. So I started from here.

[ahakanbaba]

From the normal mode in neovim I can run the following commands

:lua print(require("vectorcode").check())

prints true

and

:lua print(require("vectorcode").query("some query message"))

prints

Started retrieving 1 documents.
table: 0x7d5e2c8adb58
Press ENTER or type command to continue

Does this suggest the vectorcode nvim plugin is setup correctly ?

[olimorris]

What LLM are you using? Quickly scanning your post, I think it's calling vector code incorrectly.

[ahakanbaba]

I was using “o3”

[ahakanbaba]

I have tried to use "gpt4.1" instead as well.

With settings like this

opts = {
    adapters = {
        openai = function()
            return require("codecompanion.adapters").extend("openai", {
                opts = {
                    stream = true,
                },
                schema = {
                    model = {
                        default = function()
                            -- The model name from openai
                            -- return "o3"
                            return "gpt-4.1"
                        end,
                    },
                },
            })
        end,
    },
    extensions = {
        vectorcode = {
            opts = {
                add_tool = true,
                add_slash_command = true,
                tool_opts = {},
            }
        },
    }
},

And trying to trigger the vectorcode with this chat

## Me

> Context:
> - <tool>vectorcode</tool>

@vectorcode how to test tensorrt ?

This time it gives me a similar error, not exactly the same.

[ERROR] 2025-05-11 23:52:49
Error: {
  "error": {
    "message": "Invalid schema for function 'vectorcode': In context=('properties', 'options'), 'required' is required to be supplied and to be an array including every key in properties. Missing 'count'.",
    "type": "invalid_request_error",
    "param": "tools[0].function.parameters",
    "code": "invalid_function_parameters"
  }
}

The previous one was missing project_root, this one is missing count

[olimorris]

Maybe post this in the vectorcode repo.

[ahakanbaba]

Thank you very much. I will follow-up with that.

Do you mind sharing just a little bit context, why do you see this error originating from that level ?

Do you anticipate something may be missing with the tool implementation, would that be here ?

[olimorris]

CodeCompanion executes the function call. The tool is responsible for defining the schema which sets out how to call the function. So if there's an error with the schema, that's an error with how the LLM is calling it how the schema has been defined

[blankenshipz]

I've also been getting this error a lot @ahakanbaba please let us know if you open an issue on the vectorcode repo

[ahakanbaba]

@blankenshipz you beat me to it. Thank you very much for creating this ticket Davidyz/VectorCode#133

[ahakanbaba]

@olimorris A relevant question for code-companion is that can we see the OpenAI responses in the logs ?

Looking into the DEBUG level information , we see this request went to openai.

Is there a way to see in the logs or otherwise, what openai responded to that request ?

[ahakanbaba]

I think enabling TRACE logging gets you see the LLM responses.
I think I got what I needed from here Davidyz/VectorCode#133 (comment)

[Davidyz]

Should've been fixed in the latest release of VectorCode (0.6.8 at the moment). Different LLM providers (not to be confused with codecompanion adapters) seem to implement tool calling slightly differently, which made it difficult for me to test the changes.

[blankenshipz]

Thank you!!!