Adding groq, openrouter and deepseek to ADAPTER list: Groq fixed!

[richtong]

OK, another noob question. Thanks to all of you I have deepseek running with codecompanion because it emulates the openai API. There is another provider groq which does the same thing, so I tried the same trick, just use the openai adapter and give it a different "utl".

The problem is that the groq emulation of openai does not accept the "id" tag. See https://console.groq.com/docs/api-reference#chat-create and suggests using a "name" tag instead. I'm looking at the Adapter code and can't for the life of me figure out how to suppress the "id" part of the JSON message.

Is there an easy way to do this, just as you can just extend the function with "URL" to get the right endpoint:

return {
  {
    "olimorris/codecompanion.nvim",
    dependencies = {
      -- { "nvim-treesitter/nvim-treesitter", build = ":TSUpdate" },
      "nvim-lua/plenary.nvim",
      "nvim-treesitter/nvim-treesitter",
      "hrsh7th/nvim-cmp",
      "nvim-telescope/telescope.nvim",
      { "stevearc/dressing.nvim", opts = {} },
    },
    config = true,
    opts = {
      adapters = {
        -- because groq is also openai api compliant
        groq = function()
          return require("codecompanion.adapters").extend("openai", {
            env = {
              api_key = vim.env.GROQ_API_KEY,
            },
            name = "Groq",
            url = "https://api.groq.com/openai/v1/chat/completions",
            schema = {
              model = {
                default = "llama3.2-11-text-preview",
              },
            },
            max_tokens = {
              default = 8192,
            },
            temperature = {
              default = 1,
            },
          })
        end,
        deepseek25 = function()
          return require("codecompanion.adapters").extend("openai", {
            name = "deepseek 2.5",
            env = {
              api_key = vim.env.DEEPSEEK_API_KEY,
            },

            url = "https://api.deepseek.com/chat/completions",
            schema = {
              model = {
                default = "deepseek-chat",
              },
            max_tokens = {
              default = 8192,
            },
            temperature = {
              default = 1,
            },
          })
        end,
     },
  },
}

[bassamsdata]

Alright, I was testing the new meta models with groq and managed to solve the issue, though the solution is a bit hacky—it works for now.

I added a handler to modify the ID. However, I believe the model name in your version might not be entirely correct. Could you take a look at it, please?

groq = function()
  return require("codecompanion.adapters").extend("openai", {
    env = {
      api_key = vim.env.GROQ_API_KEY,
    },
    name = "Groq",
    url = "https://api.groq.com/openai/v1/chat/completions",
    schema = {
      model = {
        default = "llama-3.2-11b-text-preview",
      },
    },
    max_tokens = {
      default = 8192,
    },
    temperature = {
      default = 1,
    },
    handlers = {
      form_messages = function(self, messages)
        for i, msg in ipairs(messages) do
          -- Remove 'id' and 'opts' properties from all messages
          msg.id = nil
          msg.opts = nil

          -- Ensure 'name' is a string if present, otherwise remove it
          if msg.name then
            msg.name = tostring(msg.name)
          else
            msg.name = nil
          end

          -- Ensure only supported properties are present
          local supported_props =
            { role = true, content = true, name = true }
          for prop in pairs(msg) do
            if not supported_props[prop] then
              msg[prop] = nil
            end
          end
        end
        return { messages = messages }
      end,
    },
  })
end,

edit: correct the code indentation
edit2: revert the API as the original OP

[olimorris]

Yes I'm very much down a PR to add these. My only request would be that you also add corresponding tests and you stick to the interface that I'm using for the existing plugins.

The ADAPTERS.md file should be a helpful resource and my tests are very simple. You just need to capture the output from the logs to ensure you're dealing with the correct response from the endpoint.

If they're all following the OpenAI standard of endpoints then even better as we can just reuse a lot of that logic. I did this for the Co-Pilot adapter.

[richtong]

That makes sense, I am happy to help, your project is AMAZING! I will work on the tests as well that is a good idea. The test set should be the same as the Co-Pilot adapter and will follow that as a model.

[richtong]

fyi @olimorris I'm working on this, too many day job things, but the work to add tests and so forth is so smart!

My main other challenge is figuring our mini.diff, so many little things

[olimorris]

fyi @olimorris I'm working on this, too many day job things, but the work to add tests and so forth is so smart!

My main other challenge is figuring our mini.diff, so many little things

That's so great to hear 😀. And I know the feeling regarding the day job.

Let me know what you're trying to do with mini.diff. My dotfiles should show you how you can set it up

[richtong]

@olimorris so I read the ADAPTERS.md, but I don't see the section on how to generate tests, is this in another document.

Ah yes, where are your dotfiles? :-)

[ezechukwu69]

Made a PR to acheive support for any OpenAI compatible model, self hosted or otherwise.
This could help, it extends openAI config and allow more customization.
My use case was usinfg LMStudio because I could not get Ollama to work with my Intel GPU
#329
@richtong

[ezechukwu69]

@olimorris PR for this feature created, using similar interface but with more customizations, noticed that this has options /v1/chat/completions and /chat/completions. This is also factored in