Add a support for Grok-4 from x.ai

README.md

@@ -131,6 +131,7 @@ The following LLM providers are currently supported besides OpenAI:
 - [Perplexity.ai](https://www.perplexity.ai/pro) Pro users have $5/month free API credits available (the default PPLX agent uses Mixtral-8x7b).
 - [Anthropic](https://www.anthropic.com/api) to access Claude models, which currently outperform GPT-4 in some benchmarks.
 - [Google Gemini](https://ai.google.dev/) with a quite generous free range but some geo-restrictions (EU).
+- [x.ai](https://x.ai/api) to access Grok-4 with pre-paid credit plans (No free credit available at this moment).
 - Any other "OpenAI chat/completions" compatible endpoint (Azure, LM Studio, etc.)
 
 Below is an example of the relevant configuration part enabling some of these. The `secret` field has the same capabilities as `openai_api_key` (which is still supported for compatibility).
@@ -171,6 +172,11 @@ Below is an example of the relevant configuration part enabling some of these. T
 			endpoint = "https://api.anthropic.com/v1/messages",
 			secret = os.getenv("ANTHROPIC_API_KEY"),
 		},
+
+		xai = {
+			endpoint = "https://api.x.ai/v1/chat/completions",
+			secret = os.getenv("XAI_API_KEY"),
+		},
 	},
 ```
 

lua/gp/config.lua

@@ -72,6 +72,11 @@ local config = {
 			endpoint = "https://api.anthropic.com/v1/messages",
 			secret = os.getenv("ANTHROPIC_API_KEY"),
 		},
+		xai = {
+			disable = true,
+			endpoint = "https://api.x.ai/v1/chat/completions",
+			secret = os.getenv("XAI_API_KEY"),
+		},
 	},
 
 	-- prefix for all commands
@@ -310,6 +315,15 @@ local config = {
 			model = { model = "claude-3-5-haiku-latest", temperature = 0.8, top_p = 1 },
 			system_prompt = require("gp.defaults").code_system_prompt,
 		},
+		{
+			provider = "xai",
+			name = "Grok-4",
+			chat = false,
+			command = true,
+			-- string with model name or table with model name and parameters
+			model = { model = "grok-4-latest", temperature = 0 },
+			system_prompt = require("gp.defaults").code_system_prompt,
+		},
 		{
 			provider = "ollama",
 			name = "CodeOllamaLlama3.1-8B",

lua/gp/dispatcher.lua

@@ -169,6 +169,30 @@ D.prepare_payload = function(messages, model, provider)
 		return payload
 	end
 
+	if provider == "xai" then
+		local system = ""
+		local i = 1
+		while i < #messages do
+			if messages[i].role == "system" then
+				system = system .. messages[i].content .. "\n"
+				table.remove(messages, i)
+			else
+				i = i + 1
+			end
+		end
+
+		local payload = {
+			model = model.model,
+			stream = true,
+			messages = messages,
+			system = system,
+			max_tokens = model.max_tokens or 4096,
+			temperature = math.max(0, math.min(2, model.temperature or 1)),
+			top_p = math.max(0, math.min(1, model.top_p or 1)),
+		}
+		return payload
+	end
+
 	if provider == "ollama" then
 		local payload = {
 			model = model.model,
@@ -452,6 +476,14 @@ local query = function(buf, provider, payload, handler, on_exit, callback)
 			"api-key: " .. bearer,
 		}
 		endpoint = render.template_replace(endpoint, "{{model}}", payload.model)
+	elseif provider == "xai" then
+		-- currently xai only uses bearer token for authentication.
+		-- since I cannot sure its going to be that way for long time
+		-- branching out as another condition.
+		headers = {
+			"-H",
+			"Authorization: Bearer " .. bearer,
+		}
 	elseif provider == "ollama" then
 		headers = {}
 	else -- default to openai compatible headers