@oh-my-pi/pi-ai 13.3.0 → 13.3.2

package/CHANGELOG.md

@@ -2,6 +2,11 @@
 
 ## [Unreleased]
 
+## [13.3.1] - 2026-02-26
+### Added
+
+- Added `topP`, `topK`, `minP`, `presencePenalty`, and `repetitionPenalty` options to `StreamOptions` for fine-grained control over model sampling behavior
+
 ## [13.3.0] - 2026-02-26
 
 ### Changed

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-ai",
-	"version": "13.3.0",
+	"version": "13.3.2",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"homepage": "https://github.com/can1357/oh-my-pi",
 	"author": "Can Boluk",
@@ -44,7 +44,7 @@
 		"@connectrpc/connect-node": "^2.1",
 		"@google/genai": "^1.42",
 		"@mistralai/mistralai": "^1.14",
-		"@oh-my-pi/pi-utils": "13.3.0",
+		"@oh-my-pi/pi-utils": "13.3.2",
 		"@sinclair/typebox": "^0.34",
 		"@smithy/node-http-handler": "^4.4",
 		"ajv": "^8.18",

package/src/providers/amazon-bedrock.ts

@@ -126,7 +126,7 @@ export const streamBedrock: StreamFunction<"bedrock-converse-stream"> = (
 				modelId: model.id,
 				messages: convertMessages(context, model, cacheRetention),
 				system: buildSystemPrompt(context.systemPrompt, model, cacheRetention),
-				inferenceConfig: { maxTokens: options.maxTokens, temperature: options.temperature },
+				inferenceConfig: { maxTokens: options.maxTokens, temperature: options.temperature, topP: options.topP },
 				toolConfig: convertToolConfig(context.tools, options.toolChoice),
 				additionalModelRequestFields: buildAdditionalModelRequestFields(model, options),
 			};

package/src/providers/anthropic.ts

@@ -111,6 +111,11 @@ export function buildAnthropicHeaders(options: AnthropicHeaderOptions): Record<s
 }
 
 type AnthropicCacheControl = { type: "ephemeral"; ttl?: "1h" | "5m" };
+
+type AnthropicSamplingParams = MessageCreateParamsStreaming & {
+	top_p?: number;
+	top_k?: number;
+};
 function getCacheControl(
 	baseUrl: string,
 	cacheRetention?: CacheRetention,
@@ -875,7 +880,7 @@ function buildParams(
 	options?: AnthropicOptions,
 ): MessageCreateParamsStreaming {
 	const { cacheControl } = getCacheControl(model.baseUrl, options?.cacheRetention);
-	const params: MessageCreateParamsStreaming = {
+	const params: AnthropicSamplingParams = {
 		model: model.id,
 		messages: convertAnthropicMessages(context.messages, model, isOAuthToken),
 		max_tokens: options?.maxTokens || (model.maxTokens / 3) | 0,
@@ -911,6 +916,12 @@ function buildParams(
 	if (options?.temperature !== undefined) {
 		params.temperature = options.temperature;
 	}
+	if (options?.topP !== undefined) {
+		params.top_p = options.topP;
+	}
+	if (options?.topK !== undefined) {
+		params.top_k = options.topK;
+	}
 
 	if (context.tools) {
 		params.tools = convertTools(context.tools, isOAuthToken);

package/src/providers/azure-openai-responses.ts

@@ -71,6 +71,14 @@ export interface AzureOpenAIResponsesOptions extends StreamOptions {
 	toolChoice?: ToolChoice;
 }
 
+type AzureOpenAIResponsesSamplingParams = ResponseCreateParamsStreaming & {
+	top_p?: number;
+	top_k?: number;
+	min_p?: number;
+	presence_penalty?: number;
+	repetition_penalty?: number;
+};
+
 /**
  * Generate function for Azure OpenAI Responses API
  */
@@ -448,7 +456,7 @@ function buildParams(
 ) {
 	const messages = convertMessages(model, context, true);
 
-	const params: ResponseCreateParamsStreaming = {
+	const params: AzureOpenAIResponsesSamplingParams = {
 		model: deploymentName,
 		input: messages,
 		stream: true,
@@ -462,6 +470,21 @@ function buildParams(
 	if (options?.temperature !== undefined) {
 		params.temperature = options?.temperature;
 	}
+	if (options?.topP !== undefined) {
+		params.top_p = options.topP;
+	}
+	if (options?.topK !== undefined) {
+		params.top_k = options.topK;
+	}
+	if (options?.minP !== undefined) {
+		params.min_p = options.minP;
+	}
+	if (options?.presencePenalty !== undefined) {
+		params.presence_penalty = options.presencePenalty;
+	}
+	if (options?.repetitionPenalty !== undefined) {
+		params.repetition_penalty = options.repetitionPenalty;
+	}
 
 	if (context.tools) {
 		params.tools = convertTools(context.tools);

package/src/providers/gitlab-duo.ts

@@ -281,6 +281,11 @@ export function streamGitLabDuo(
 								apiKey: directAccess.token,
 								isOAuth: true,
 								temperature: options.temperature,
+								topP: options.topP,
+								topK: options.topK,
+								minP: options.minP,
+								presencePenalty: options.presencePenalty,
+								repetitionPenalty: options.repetitionPenalty,
 								maxTokens: options.maxTokens ?? Math.min(model.maxTokens, 32000),
 								signal: options.signal,
 								cacheRetention: options.cacheRetention,
@@ -310,6 +315,11 @@ export function streamGitLabDuo(
 								{
 									apiKey: directAccess.token,
 									temperature: options.temperature,
+									topP: options.topP,
+									topK: options.topK,
+									minP: options.minP,
+									presencePenalty: options.presencePenalty,
+									repetitionPenalty: options.repetitionPenalty,
 									maxTokens: options.maxTokens ?? model.maxTokens,
 									signal: options.signal,
 									cacheRetention: options.cacheRetention,
@@ -334,6 +344,11 @@ export function streamGitLabDuo(
 								{
 									apiKey: directAccess.token,
 									temperature: options.temperature,
+									topP: options.topP,
+									topK: options.topK,
+									minP: options.minP,
+									presencePenalty: options.presencePenalty,
+									repetitionPenalty: options.repetitionPenalty,
 									maxTokens: options.maxTokens ?? model.maxTokens,
 									signal: options.signal,
 									cacheRetention: options.cacheRetention,

package/src/providers/google-gemini-cli.ts

@@ -268,6 +268,11 @@ interface CloudCodeAssistRequest {
 		generationConfig?: {
 			maxOutputTokens?: number;
 			temperature?: number;
+			topP?: number;
+			topK?: number;
+			minP?: number;
+			presencePenalty?: number;
+			repetitionPenalty?: number;
 			thinkingConfig?: ThinkingConfig;
 		};
 		tools?: { functionDeclarations: Record<string, unknown>[] }[] | undefined;
@@ -829,6 +834,21 @@ export function buildRequest(
 	if (options.maxTokens !== undefined) {
 		generationConfig.maxOutputTokens = options.maxTokens;
 	}
+	if (options.topP !== undefined) {
+		generationConfig.topP = options.topP;
+	}
+	if (options.topK !== undefined) {
+		generationConfig.topK = options.topK;
+	}
+	if (options.minP !== undefined) {
+		generationConfig.minP = options.minP;
+	}
+	if (options.presencePenalty !== undefined) {
+		generationConfig.presencePenalty = options.presencePenalty;
+	}
+	if (options.repetitionPenalty !== undefined) {
+		generationConfig.repetitionPenalty = options.repetitionPenalty;
+	}
 
 	// Thinking config
 	if (options.thinking?.enabled && model.reasoning) {

package/src/providers/google-vertex.ts

@@ -42,6 +42,14 @@ export interface GoogleVertexOptions extends StreamOptions {
 	location?: string;
 }
 
+interface GoogleVertexSamplingConfig extends GenerateContentConfig {
+	topP?: number;
+	topK?: number;
+	minP?: number;
+	presencePenalty?: number;
+	repetitionPenalty?: number;
+}
+
 const API_VERSION = "v1";
 
 const THINKING_LEVEL_MAP: Record<GoogleThinkingLevel, ThinkingLevel> = {
@@ -338,13 +346,28 @@ function buildParams(
 ): GenerateContentParameters {
 	const contents = convertMessages(model, context);
 
-	const generationConfig: GenerateContentConfig = {};
+	const generationConfig: GoogleVertexSamplingConfig = {};
 	if (options.temperature !== undefined) {
 		generationConfig.temperature = options.temperature;
 	}
 	if (options.maxTokens !== undefined) {
 		generationConfig.maxOutputTokens = options.maxTokens;
 	}
+	if (options.topP !== undefined) {
+		generationConfig.topP = options.topP;
+	}
+	if (options.topK !== undefined) {
+		generationConfig.topK = options.topK;
+	}
+	if (options.minP !== undefined) {
+		generationConfig.minP = options.minP;
+	}
+	if (options.presencePenalty !== undefined) {
+		generationConfig.presencePenalty = options.presencePenalty;
+	}
+	if (options.repetitionPenalty !== undefined) {
+		generationConfig.repetitionPenalty = options.repetitionPenalty;
+	}
 
 	const config: GenerateContentConfig = {
 		...(Object.keys(generationConfig).length > 0 && generationConfig),

package/src/providers/google.ts

@@ -42,6 +42,14 @@ export interface GoogleOptions extends StreamOptions {
 	};
 }
 
+interface GoogleSamplingConfig extends GenerateContentConfig {
+	topP?: number;
+	topK?: number;
+	minP?: number;
+	presencePenalty?: number;
+	repetitionPenalty?: number;
+}
+
 // Counter for generating unique tool call IDs
 let toolCallCounter = 0;
 
@@ -304,13 +312,28 @@ function buildParams(
 ): GenerateContentParameters {
 	const contents = convertMessages(model, context);
 
-	const generationConfig: GenerateContentConfig = {};
+	const generationConfig: GoogleSamplingConfig = {};
 	if (options.temperature !== undefined) {
 		generationConfig.temperature = options.temperature;
 	}
 	if (options.maxTokens !== undefined) {
 		generationConfig.maxOutputTokens = options.maxTokens;
 	}
+	if (options.topP !== undefined) {
+		generationConfig.topP = options.topP;
+	}
+	if (options.topK !== undefined) {
+		generationConfig.topK = options.topK;
+	}
+	if (options.minP !== undefined) {
+		generationConfig.minP = options.minP;
+	}
+	if (options.presencePenalty !== undefined) {
+		generationConfig.presencePenalty = options.presencePenalty;
+	}
+	if (options.repetitionPenalty !== undefined) {
+		generationConfig.repetitionPenalty = options.repetitionPenalty;
+	}
 
 	const config: GenerateContentConfig = {
 		...(Object.keys(generationConfig).length > 0 && generationConfig),

package/src/providers/kimi.ts

@@ -70,6 +70,11 @@ export function streamKimi(
 				const innerStream = streamAnthropic(anthropicModel, context, {
 					apiKey: options?.apiKey,
 					temperature: options?.temperature,
+					topP: options?.topP,
+					topK: options?.topK,
+					minP: options?.minP,
+					presencePenalty: options?.presencePenalty,
+					repetitionPenalty: options?.repetitionPenalty,
 					maxTokens: options?.maxTokens ?? Math.min(model.maxTokens, 32000),
 					signal: options?.signal,
 					headers: mergedHeaders,
@@ -87,6 +92,11 @@ export function streamKimi(
 				const innerStream = streamOpenAICompletions(model, context, {
 					apiKey: options?.apiKey,
 					temperature: options?.temperature,
+					topP: options?.topP,
+					topK: options?.topK,
+					minP: options?.minP,
+					presencePenalty: options?.presencePenalty,
+					repetitionPenalty: options?.repetitionPenalty,
 					maxTokens: options?.maxTokens ?? model.maxTokens,
 					signal: options?.signal,
 					headers: mergedHeaders,

package/src/providers/openai-codex/request-transformer.ts

@@ -30,6 +30,11 @@ export interface RequestBody {
 	tools?: unknown;
 	tool_choice?: unknown;
 	temperature?: number;
+	top_p?: number;
+	top_k?: number;
+	min_p?: number;
+	presence_penalty?: number;
+	repetition_penalty?: number;
 	reasoning?: Partial<ReasoningConfig>;
 	text?: {
 		verbosity?: "low" | "medium" | "high";

package/src/providers/openai-codex-responses.ts

@@ -333,6 +333,21 @@ export const streamOpenAICodexResponses: StreamFunction<"openai-codex-responses"
 			if (options?.temperature !== undefined) {
 				params.temperature = options.temperature;
 			}
+			if (options?.topP !== undefined) {
+				params.top_p = options.topP;
+			}
+			if (options?.topK !== undefined) {
+				params.top_k = options.topK;
+			}
+			if (options?.minP !== undefined) {
+				params.min_p = options.minP;
+			}
+			if (options?.presencePenalty !== undefined) {
+				params.presence_penalty = options.presencePenalty;
+			}
+			if (options?.repetitionPenalty !== undefined) {
+				params.repetition_penalty = options.repetitionPenalty;
+			}
 
 			if (context.tools && context.tools.length > 0) {
 				params.tools = convertTools(context.tools);

package/src/providers/openai-completions.ts

@@ -109,6 +109,12 @@ export interface OpenAICompletionsOptions extends StreamOptions {
 	reasoningEffort?: "minimal" | "low" | "medium" | "high" | "xhigh";
 }
 
+type OpenAICompletionsSamplingParams = OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming & {
+	top_k?: number;
+	min_p?: number;
+	repetition_penalty?: number;
+};
+
 // LIMITATION: The think tag parser uses naive string matching for <think>/<thinking> tags.
 // If MiniMax models output these literal strings in code blocks, XML examples, or explanations,
 // they will be incorrectly consumed as thinking delimiters, truncating visible output.
@@ -530,7 +536,7 @@ function buildParams(model: Model<"openai-completions">, context: Context, optio
 	const isKimi = model.id.includes("moonshotai/kimi");
 	const effectiveMaxTokens = options?.maxTokens ?? (isKimi ? model.maxTokens : undefined);
 
-	const params: OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming = {
+	const params: OpenAICompletionsSamplingParams = {
 		model: model.id,
 		messages,
 		stream: true,
@@ -555,6 +561,21 @@ function buildParams(model: Model<"openai-completions">, context: Context, optio
 	if (options?.temperature !== undefined) {
 		params.temperature = options.temperature;
 	}
+	if (options?.topP !== undefined) {
+		params.top_p = options.topP;
+	}
+	if (options?.topK !== undefined) {
+		params.top_k = options.topK;
+	}
+	if (options?.minP !== undefined) {
+		params.min_p = options.minP;
+	}
+	if (options?.presencePenalty !== undefined) {
+		params.presence_penalty = options.presencePenalty;
+	}
+	if (options?.repetitionPenalty !== undefined) {
+		params.repetition_penalty = options.repetitionPenalty;
+	}
 
 	if (context.tools) {
 		params.tools = convertTools(context.tools, compat);

package/src/providers/openai-responses.ts

@@ -65,6 +65,14 @@ export interface OpenAIResponsesOptions extends StreamOptions {
 	strictResponsesPairing?: boolean;
 }
 
+type OpenAIResponsesSamplingParams = ResponseCreateParamsStreaming & {
+	top_p?: number;
+	top_k?: number;
+	min_p?: number;
+	presence_penalty?: number;
+	repetition_penalty?: number;
+};
+
 /**
  * Generate function for OpenAI Responses API
  */
@@ -407,7 +415,7 @@ function buildParams(model: Model<"openai-responses">, context: Context, options
 
 	const cacheRetention = resolveCacheRetention(options?.cacheRetention);
 	const promptCacheKey = cacheRetention === "none" ? undefined : options?.sessionId;
-	const params: ResponseCreateParamsStreaming = {
+	const params: OpenAIResponsesSamplingParams = {
 		model: model.id,
 		input: messages,
 		stream: true,
@@ -423,6 +431,21 @@ function buildParams(model: Model<"openai-responses">, context: Context, options
 	if (options?.temperature !== undefined) {
 		params.temperature = options?.temperature;
 	}
+	if (options?.topP !== undefined) {
+		params.top_p = options.topP;
+	}
+	if (options?.topK !== undefined) {
+		params.top_k = options.topK;
+	}
+	if (options?.minP !== undefined) {
+		params.min_p = options.minP;
+	}
+	if (options?.presencePenalty !== undefined) {
+		params.presence_penalty = options.presencePenalty;
+	}
+	if (options?.repetitionPenalty !== undefined) {
+		params.repetition_penalty = options.repetitionPenalty;
+	}
 
 	if (options?.serviceTier !== undefined) {
 		params.service_tier = options.serviceTier;

package/src/providers/synthetic.ts

@@ -67,6 +67,11 @@ export function streamSynthetic(
 				const innerStream = streamAnthropic(anthropicModel, context, {
 					apiKey: options?.apiKey,
 					temperature: options?.temperature,
+					topP: options?.topP,
+					topK: options?.topK,
+					minP: options?.minP,
+					presencePenalty: options?.presencePenalty,
+					repetitionPenalty: options?.repetitionPenalty,
 					maxTokens: options?.maxTokens ?? Math.min(model.maxTokens, 32000),
 					signal: options?.signal,
 					headers: mergedHeaders,
@@ -90,6 +95,11 @@ export function streamSynthetic(
 				const innerStream = streamOpenAICompletions(syntheticModel, context, {
 					apiKey: options?.apiKey,
 					temperature: options?.temperature,
+					topP: options?.topP,
+					topK: options?.topK,
+					minP: options?.minP,
+					presencePenalty: options?.presencePenalty,
+					repetitionPenalty: options?.repetitionPenalty,
 					maxTokens: options?.maxTokens ?? model.maxTokens,
 					signal: options?.signal,
 					headers: mergedHeaders,

package/src/stream.ts

@@ -396,6 +396,11 @@ function mapOptionsForApi<TApi extends Api>(
 ): OptionsForApi<TApi> {
 	const base = {
 		temperature: options?.temperature,
+		topP: options?.topP,
+		topK: options?.topK,
+		minP: options?.minP,
+		presencePenalty: options?.presencePenalty,
+		repetitionPenalty: options?.repetitionPenalty,
 		maxTokens: options?.maxTokens || Math.min(model.maxTokens, 32000),
 		signal: options?.signal,
 		apiKey: apiKey || options?.apiKey,

package/src/types.ts

@@ -129,6 +129,11 @@ export interface ProviderSessionState {
 
 export interface StreamOptions {
 	temperature?: number;
+	topP?: number;
+	topK?: number;
+	minP?: number;
+	presencePenalty?: number;
+	repetitionPenalty?: number;
 	maxTokens?: number;
 	signal?: AbortSignal;
 	apiKey?: string;