bug: API request error status 422 deserialize JSON missing field parameters

[paulrougieux]

Describe the bug

xai model = "grok-code-fast-1" used to work in October 20205. Since version 216ba4a , when I use avante, I see an API request error:

Error: API request failed with status 422. Body: "Failed
to deserialize the JSON body into the target type:
tools[8]: missing field `parameters` at line 1 column
6488"

Reverting to a previous version fixes the issue, so it's not an API issue. More details below.

Related issues

An issue search for API request status 422 deserialize JSON missing field parameters returns one issue from a year ago:

• bug: Can not submit question #422 one comment suggests downgrading avante to a previous commit

To reproduce

Activate avante with leader key aa or :AvanteAsk, then ask a question in the chat window. The error will appear with current version of avante as ov Nov 7 2025. I am using grok code fast as the default provider, as you can see in the configuration below and also in the reproducible example in repro.lua.

Expected behavior

The use of :AvanteAsk should work without returning an API error.

Temporary fix

1. Reverting to a previous commit from 3 weeks ago (October 15, 2025) in nvim/lua/plugins/avante.lua fixes the issue:

   "yetone/avante.nvim",
   ... other lines ...
   commit = "216ba4a8b2b75", -- Revert to a previous commit if needed

Installation method

return {
  {
    "yetone/avante.nvim",
    -- if you want to build from source then do `make BUILD_FROM_SOURCE=true`
    -- ⚠️ must add this setting! ! !
    build = vim.fn.has("win32") ~= 0
        and "powershell -ExecutionPolicy Bypass -File Build.ps1 -BuildFromSource false"
        or "make",
    event = "VeryLazy",
    version = false, -- Never set this value to "*"! Never!
    commit = "216ba4a8b2b75", -- Revert to a previous commit if needed
    ---@module 'avante'
    ---@type avante.Config
    opts = {
      -- add any opts here
      -- this file can contain specific instructions for your project
      instructions_file = "avante.md",
      -- Set xai as the default provider
      provider = "xai",
      -- Options for each provider
      providers = {
        claude = {
          endpoint = "https://api.anthropic.com",
          model = "claude-sonnet-4-20250514",
          timeout = 30000, -- Timeout in milliseconds
            extra_request_body = {
              temperature = 0.75,
              max_tokens = 20480,
            },
        },
        xai = {
          model = "grok-code-fast-1", -- Or "grok-4" / keep as  for speed
          temperature = 0.2,
          max_tokens = 4096,
        },
      },
    },
    dependencies = {
      "nvim-lua/plenary.nvim",
      "MunifTanjim/nui.nvim",
      --- The below dependencies are optional,
      "nvim-mini/mini.pick", -- for file_selector provider mini.pick
      "nvim-telescope/telescope.nvim", -- for file_selector provider telescope
      "hrsh7th/nvim-cmp", -- autocompletion for avante commands and mentions
      "ibhagwan/fzf-lua", -- for file_selector provider fzf
      "stevearc/dressing.nvim", -- for input provider dressing
      "folke/snacks.nvim", -- for input provider snacks
      "nvim-tree/nvim-web-devicons", -- or echasnovski/mini.icons
      "zbirenbaum/copilot.lua", -- for providers='copilot'
      {
        -- support for image pasting
        "HakonHarnes/img-clip.nvim",
        event = "VeryLazy",
        opts = {
          -- recommended settings
          default = {
            embed_image_as_base64 = false,
            prompt_for_file_name = false,
            drag_and_drop = {
              insert_mode = true,
            },
            -- required for Windows users
            use_absolute_path = true,
          },
        },
      },
      {
        -- Make sure to set this up properly if you have lazy=true
        'MeanderingProgrammer/render-markdown.nvim',
        opts = {
          file_types = { "markdown", "Avante" },
        },
        ft = { "markdown", "Avante" },
      },
    },
  }
}

Environment

nvim -v
NVIM v0.10.4
Build type: Release
LuaJIT 2.1.1737090214

 uname -a
Linux castanea 6.12.48+deb13-amd64 #1 SMP PREEMPT_DYNAMIC Debian 6.12.48-1 (2025-09-20) x86_64 GNU/Linux

Repro

vim.env.LAZY_STDPATH = ".repro"
load(vim.fn.system("curl -s https://raw.githubusercontent.com/folke/lazy.nvim/main/bootstrap.lua"))()

require("lazy.minit").repro({
  spec = {
    -- add any other plugins here
      {
    "yetone/avante.nvim",
    -- if you want to build from source then do `make BUILD_FROM_SOURCE=true`
    -- ⚠️ must add this setting! ! !
    build = vim.fn.has("win32") ~= 0
        and "powershell -ExecutionPolicy Bypass -File Build.ps1 -BuildFromSource false"
        or "make",
    event = "VeryLazy",
    version = false, -- Never set this value to "*"! Never!
    -- commit = "216ba4a8b2b75", -- Revert to this previous commit to make it work again
    ---@module 'avante'
    ---@type avante.Config
    opts = {
      -- add any opts here
      -- this file can contain specific instructions for your project
      instructions_file = "avante.md",
      -- Set xai as the default provider
      provider = "xai",
      -- Options for each provider
      providers = {
        claude = {
          endpoint = "https://api.anthropic.com",
          model = "claude-sonnet-4-20250514",
          timeout = 30000, -- Timeout in milliseconds
            extra_request_body = {
              temperature = 0.75,
              max_tokens = 20480,
            },
        },
        xai = {
          model = "grok-code-fast-1", -- Or "grok-4" / keep as  for speed
          temperature = 0.2,
          max_tokens = 4096,
        },
      },
    },
    dependencies = {
      "nvim-lua/plenary.nvim",
      "MunifTanjim/nui.nvim",
      --- The below dependencies are optional,
      "nvim-mini/mini.pick", -- for file_selector provider mini.pick
      "nvim-telescope/telescope.nvim", -- for file_selector provider telescope
      "hrsh7th/nvim-cmp", -- autocompletion for avante commands and mentions
      "ibhagwan/fzf-lua", -- for file_selector provider fzf
      "stevearc/dressing.nvim", -- for input provider dressing
      "folke/snacks.nvim", -- for input provider snacks
      "nvim-tree/nvim-web-devicons", -- or echasnovski/mini.icons
      "zbirenbaum/copilot.lua", -- for providers='copilot'
      {
        -- support for image pasting
        "HakonHarnes/img-clip.nvim",
        event = "VeryLazy",
        opts = {
          -- recommended settings
          default = {
            embed_image_as_base64 = false,
            prompt_for_file_name = false,
            drag_and_drop = {
              insert_mode = true,
            },
            -- required for Windows users
            use_absolute_path = true,
          },
        },
      },
      {
        -- Make sure to set this up properly if you have lazy=true
        'MeanderingProgrammer/render-markdown.nvim',
        opts = {
          file_types = { "markdown", "Avante" },
        },
        ft = { "markdown", "Avante" },
      },
    },
  }
  },
})

[paulrougieux]

I'm wondering why this parameters field is now missing from the JSON.

Notes on the xai provider :

• The version 216ba4a from 3 weeks ago that is working doesn't contain a dedicated provider for xai. So I suppose it uses a default provider. Not sure if avante/providers is the right place to look https://github.com/yetone/avante.nvim/tree/216ba4a8b2b751e2e86da6f7d8fe59b1a67c1197/lua/avante/providers

• I see an xai provider defined in lua/avante/config.lua, it specifies __inherited_from = "openai",

• Why is the xai provider visible in this commit

  avante.nvim/lua/avante/config.lua

  Line 486 in 44b5948

  endpoint = "https://api.x.ai/v1",

  but not in the main version of lua/avante/config.lua.

Comparing between a commit from 3 weeks ago (where the bug didn't appear) and a more recent commit where the bug appears 216ba4a...b95e27b
I asked Claude to search for "parameters" in the file diffs and to explain what could have caused the change. https://claude.ai/share/e3b18f1b-9802-4266-909a-01d8bba69680 Claude response:

One of your tools (the 8th one in the array) either:

• Has parameters = nil after transformation
• Is getting the flattened structure applied incorrectly
• Has a missing or malformed parameters field in its original definition

"Quick Fix to Test: Check your XAI provider config and ensure use_response_api is explicitly set to false:"

providers = {
  xai = {
    use_response_api = false,
    --  other xai settings
  }
}

"The Response API changes were designed for OpenAI/Copilot but may not be compatible with XAI's API expectations."

• [UPDATE] use_response_api = false doesn't fix the issue

[paulrougieux]

I'm wondering if it's related to a change to the default way of handling the response API in this commit:
d45b622

# from 
use_response_api = false, -- Set to true to use OpenAI's new Response API (/responses) instead of Chat Completions API (/chat/completions)
# To
use_response_api = copilot_use_response_api, -- Automatically switch to Response API for GPT-5 Codex models

But it's probably not related since this issue arises for the grok code fast model. Although grok code fast uses the openAI api bindings, so maybe it's related?

[lmariscal]

if it's an inconsistency with the Response API on xAI side let me know and we can address it

[paulrougieux]

Above messages where only suppositions. I don't know how Avante handles API requests to grok code fast. I don't understand what changed in Avante's API handling to cause the error.

To be able to use grok code fast, I needed to switch Avante to an earlier version from a few weeks ago.

• Using certain versions of Avante with grok code fast returned the API request error mentioned in the title of this issue.

• Reverting Avante to commit = "216ba4a8b2b75", in nvim/lua/plugins/avante.lua I could use grok code fast in avante and it didn't return an error.

[UPDATED]

• Using today's version of Avante (5df39b4 latest commit in main)
  • avante.nvim 38.68ms VeryLazy ■ updated from 216ba4a to 5df39b4
    The API error doesn't appear any more.
    The day after the API error is there a gain, see my next comment below.

[lmariscal]

It seems like b95e27b added read_todos.lua which has empty fields.
This translates to no parameters being set for the tool, which is not valid in the xAI API tool call.

#2845 should fix this issue.

@paulrougieux if you could please double check.
I'm pretty sure that the latest commit 5df39b4 doesn't actually address the issue.

[gtnbssn]

I am seeing a similar error with Mistral:

Error: API request failed with status 422. Body: "{\"detail\":[{\"type\":\"missing\",\"loc\":[\"body\",\"tools\",\"list[Tool]\",8,\"function\",\"parameters\"],\"msg\":\"Field required\",\"input\":{\"description\":\"Read TODOs from the current task\",\"name\":\"read_todos\"}},{\"type\":\"string_type\",\"loc\":[\"body\",\"tools\",\"list[str]\",0],\"msg\":\"Input should be a valid string\",\"input\":{\"function\":{\"description\":\"Launch a new agent that has access to the following tools: `glob`, `grep`, `ls`, `view`, `attempt_completion`. When you are searching for a keyword or file and are not confident that you will find the right match on the first try, use the Agent tool to perform the search for you. For example:\ [...]

Could it be the same root cause?

[paulrougieux]

@lmariscal

@paulrougieux if you could please double check. I'm pretty sure that the latest commit 5df39b4 doesn't actually address the issue.

You are right I moved Avante to 5df39b4 yesterday evening, (somehow I didn't have the error yesterday evening, I did restart Neovim). This morning I have the error again when using grok code fast :

Error: API request failed with status 422. Body: 
"Failed to deserialize the JSON body into the target type: tools[8]:
 missing field `parameters` at line 1 column 6451"

What information can I provide to help you investigate?

[lmariscal]

@paulrougieux @gtnbssn
if you guys could please help verify the fix above, by either temporarily using the fork, or manually applying the fix.

Repo change it to lmariscal/avante.nvim, you can revert it after it has been merged. Or feel free to manually fork and apply the same patch.

[paulrougieux]

@lmariscal

I moved Avante to the commit hash of your fork d2b56ab?diff=split&w=0 with

   commit = "d2b56ab19b5225236a35c27ad8fd3bb10631954f", ---Revert to the commit of a pull request

and I can confirm that the "grok-code-fast-1" API request is successful.

[gtnbssn]

Did not work for me, must be a different issue with Mistral.

[paulrougieux]

#2845 has been merged and appears to fix the issue for now.

Closing.