How to connect to deepseek-v3 model?

[riggy2013]

Hi,

I used below code in init.lua (with lazy.vim as plugin manager):

  -- CodeCompanion
  {
    "olimorris/codecompanion.nvim",
    dependencies = {
      "nvim-lua/plenary.nvim",
    },
    opts = {
      adapters = {
        http = {
          ds = function()
            return require("codecompanion.adapters").extend("deepseek", {
              env = {
                api_key = "sk-xxxxxx", -- At the moment I left it hardcoded
              },
              schema = {
                model = {
                  default = "deepseek-v3",
                  choices = {
                    "deepseek-v3",
                  },
                },
              },
              url = "https://api.deepseek.com",
            })
          end,
        },
      },
      -- Configure Deepseek
      strategies = {
        chat = { adapter = "ds", },
        inline = { adapter = "ds", },
        agent = { adapter = "ds", },
      },
    },
  },

Firstly it reported error of "nil model_opts" of below code:

      if model_opts.opts and model_opts.opts.can_use_tools == false then
        if tools and vim.tbl_count(tools) > 0 then
          log:warn("Tools are not supported for this model")
        end
        return
      end
      return openai.handlers.form_tools(self, tools)

I then updated it with

      if model_opts and model_opts.opts and model_opts.opts.can_use_tools == false then

Then it report error:

[ERROR] 2025-11-27 20:28:41
Error: 
[ERROR] 2025-11-27 20:32:45
[chat::_submit_http] Error: {
  body = "",
  exit = 0,
  headers = { "Date: Thu, 27 Nov 2025 12:28:16 GMT", "Content-Length: 0", "Connection: keep-alive", "Set-Cookie: HWWAFSESTIME=1764246493161; path=/", "Set-Cookie: HWWAFSESID=24d8c94acce82c86be; path=/", "vary: origin, access-control-request-method, access-control-request-headers", "access-control-allow-credentials: true", "x-ds-trace-id: c1c299ccd2b119bb5cefc9b918db01b1", "Strict-Transport-Security: max-age=31536000; includeSubDomains; preload", "X-Content-Type-Options: nosniff", "Server: CW" },
  status = 404
}

It seemed my configuration was wrong. But I couldn't find an example of DeepSeek (either official or self-deployed) configuration. Will you please help?

Thanks,
riggy2013

[olimorris]

Why not use the default adapter and just add the API key?

[riggy2013]

How to use the default adapter with API key? Pls point me to the example. I also need to use a local deployed deepseek LLM. I encountered similar error and use official site to debug the issue.

[olimorris]

This is all in the docs under configure. Huge amount of CodeCompanion users are using DeepSeek. I'd advise searching GitHub for their config.

[riggy2013]

Hi, Oli,

I did some experiments and got below results:

1. Just used default deepseek and provided DEEPSEEK_API_KEY through env variables:

  -- CodeCompanion
  {
    "olimorris/codecompanion.nvim",
    dependencies = {
      "nvim-lua/plenary.nvim",
    },
    opts = {
      -- Configure Deepseek
      strategies = {
        chat = {
          adapter = "deepseek",
        },
      },
    },
  },

This worked with deepseek and its default model was deepseek-r1.

2. I tried to change the model to deepseek-v3 (I have to, as my local model is this one). I read the doc and the only way is redefine the adapter.

  -- CodeCompanion
  {
    "olimorris/codecompanion.nvim",
    dependencies = {
      "nvim-lua/plenary.nvim",
    },
    opts = {
      adapters = {
        http = {
          deepseek = function()
            return require("codecompanion.adapters").extend("deepseek", {
              schema = {
                model = {
                  default = "deepseek-v3",
                },
              },
            })
          end,
        },
      },
      -- Configure Deepseek
      strategies = {
        chat = {
          adapter = "deepseek",
        },
      },
    },
  },

However, this didn't work. it reported below error:

[ERROR] 2025-11-28 10:51:53
Error: {"error":{"message":"Model Not Exist","type":"invalid_request_error","param":null,"code":"invalid_request_error"}}
[ERROR] 2025-11-28 10:51:53
[chat::_submit_http] Error: {
  body = '{"error":{"message":"Model Not Exist","type":"invalid_request_error","param":null,"code":"invalid_request_error"}}',
  exit = 0,
  headers = { "Date: Fri, 28 Nov 2025 02:50:02 GMT", "Content-Type: application/json", "Content-Length: 114", "Connection: keep-alive", "Set-Cookie: HWWAFSESTIME=1764298200190; path=/", "Set-Cookie: HWWAFSESID=85864ac129a57e2fdf; path=/", "vary: origin, access-control-request-method, access-control-request-headers", "access-control-allow-credentials: true", "x-ds-trace-id: 9cef5b8279a829a98fc727a0fa0008c2", "Strict-Transport-Security: max-age=31536000; includeSubDomains; preload", "X-Content-Type-Options: nosniff", "Server: CW" },
  status = 400

My debug code shows model deepseek-v3 but model-opts is nil.

3. I checked deepseek website and found their model name is actually "deepseek-chat". So I updated as well.

              schema = {
                model = {
                  default = "deepseek-chat",
                },
              },

And now it worked with official deepseek.

4. I changed the settings to my local deepseek and model name.

              env = {
                url = "<my local model url>",
                api_key = "sk-xxxx",
	            },
              schema = {
                model = {
                  default = "deepseek-v3",
                },
              },

But now, it again showed model deepseek-v3 and model_opts nil and reported above error.

5. I used my above code workaround of this model_opts nil issue. And now it reported authentication failure.

[ERROR] 2025-11-28 13:15:23
Error: {"error":{"message":"Authentication Fails, Your api key: ****3f80 is invalid","type":"authentication_error","param":null,"code":"invalid_request_error"}}
[ERROR] 2025-11-28 13:15:39
[chat::_submit_http] Error: {
  body = '{"error":{"message":"Authentication Fails, Your api key: ****3f80 is invalid","type":"authentication_error","param":null,"code":"invalid_request_error"}}',
  exit = 0,
  headers = { "Date: Fri, 28 Nov 2025 05:15:22 GMT", "Content-Type: application/json", "Content-Length: 153", "Connection: keep-alive", "Set-Cookie: HWWAFSESTIME=1764306919502; path=/", "Set-Cookie: HWWAFSESID=02294af0c124e70440; path=/", "vary: origin, access-control-request-method, access-control-request-headers", "access-control-allow-credentials: true", "x-ds-trace-id: 661250bc096b185657f1552bd1fa2edc", "Strict-Transport-Security: max-age=31536000; includeSubDomains; preload", "X-Content-Type-Options: nosniff", "Server: CW" },
  status = 401
}

I wrote a simple python test script with same url/model name/api key and it passed.

I am a little bit confused. Is this "model.default" a key from model provider? or a key in codecompanion?

Thanks,
David

[olimorris]

I'm not sure if there's an ask of me here?! As I've stated before - read the docs. Everything you need to do with an adapter is there. If you're seeing an error, either you've messed up the config or you've entered something that DeepSeek doesn't like. There's a lot of public config's using this adapter.

[riggy2013]

Hi, Oli,

Thanks for referring me to the public config's. I tried to search with "codecompanion deepseek".

There are two issues with my config. The url should be a complete one with "v1/chat/completions". The choices should be

            return require("codecompanion.adapters").extend("deepseek", {
              url = "http://<my local deepseek>/v1/chat/completions", -- need "v1/chat/completions"
              env = {
                api_key = "sk-xxxx",
	            },
              schema = {
                model = {
                  default = "deepseek-v3",
                  choices = {
                    ["deepseek-v3"] = {}, -- need an empty for a hash.
                  },
                },
              },
            })

For the second issue, I feel it is better to change the deepseek.lua code

-- current code
      if model_opts.opts and model_opts.opts.can_use_tools == false then 
-- suggested code
     if model_opts and model_opts.opts and model_opts.opts.can_use_tools == false then

Anyway, my issue got resolved.
Thanks,
riggy2013