feat: add native Ollama API support with "think" control (#277)

* add support for ollama's direct /api/chat processing

* add support for think and options parameters

* add 'ChatQwen3-8B' as default agent (thinking-disabled qwen model)

* remove comment

* clarify comment

Author: jordibeen

lua/gp/config.lua

@@ -49,7 +49,7 @@ local config = {
 		},
 		ollama = {
 			disable = true,
-			endpoint = "http://localhost:11434/v1/chat/completions",
+			endpoint = "http://localhost:11434/api/chat",
 			secret = "dummy_secret",
 		},
 		lmstudio = {
@@ -206,6 +206,19 @@ local config = {
 			-- system prompt (use this to specify the persona/role of the AI)
 			system_prompt = "You are a general AI assistant.",
 		},
+		{
+			provider = "ollama",
+			name = "ChatQwen3-8B",
+			chat = true,
+			command = false,
+			-- string with model name or table with model name and parameters
+			model = {
+				model = "qwen3:8b",
+				think = false, -- toggle thinking mode for Ollama's thinking models
+			},
+			-- system prompt (use this to specify the persona/role of the AI)
+			system_prompt = "You are a general AI assistant.",
+		},
 		{
 			provider = "lmstudio",
 			name = "ChatLMStudio",

lua/gp/dispatcher.lua

@@ -169,6 +169,41 @@ D.prepare_payload = function(messages, model, provider)
 		return payload
 	end
 
+	if provider == "ollama" then
+		local payload = {
+			model = model.model,
+			stream = true,
+			messages = messages,
+		}
+
+		if model.think ~= nil then
+			payload.think = model.think
+		end
+
+		local options = {}
+		if model.temperature then
+			options.temperature = math.max(0, math.min(2, model.temperature))
+		end
+		if model.top_p then
+			options.top_p = math.max(0, math.min(1, model.top_p))
+		end
+		if model.min_p then
+			options.min_p = math.max(0, math.min(1, model.min_p))
+		end
+		if model.num_ctx then
+			options.num_ctx = model.num_ctx
+		end
+		if model.top_k then
+			options.top_k = model.top_k
+		end
+
+		if next(options) then
+			payload.options = options
+		end
+
+		return payload
+	end
+
 	local output = {
 		model = model.model,
 		stream = true,
@@ -289,6 +324,15 @@ local query = function(buf, provider, payload, handler, on_exit, callback)
 					end
 				end
 
+				if qt.provider == "ollama" then
+					if line:match('"message":') and line:match('"content":') then
+						local success, decoded = pcall(vim.json.decode, line)
+						if success and decoded.message and decoded.message.content then
+							content = decoded.message.content
+						end
+					end
+				end
+
 
 				if content and type(content) == "string" then
 					qt.response = qt.response .. content
@@ -408,6 +452,8 @@ local query = function(buf, provider, payload, handler, on_exit, callback)
 			"api-key: " .. bearer,
 		}
 		endpoint = render.template_replace(endpoint, "{{model}}", payload.model)
+	elseif provider == "ollama" then
+		headers = {}
 	else -- default to openai compatible headers
 		headers = {
 			"-H",