@oh-my-pi/pi-ai 15.0.0 → 15.0.2

package/CHANGELOG.md

@@ -2,6 +2,22 @@
 
 ## [Unreleased]
 
+## [15.0.2] - 2026-05-15
+### Fixed
+
+- Fixed `StreamOptions.fetch` typing to accept fetch-compatible override functions that do not expose `preconnect`, allowing custom fetch implementations to be used without type errors across runtimes
+- Fixed Moonshot Kimi K2.6 forced tool calls to send `thinking: { type: "disabled" }`, avoiding `tool_choice 'specified' is incompatible with thinking enabled` 400s while preserving the requested named tool ([#1077](https://github.com/can1357/oh-my-pi/issues/1077)).
+
+## [15.0.1] - 2026-05-14
+### Breaking Changes
+
+- Increased the minimum Bun runtime version to `>=1.3.14` for the `@aws-?` package
+
+### Added
+
+- Added `installH2Fetch` to patch `globalThis.fetch` so HTTPS requests attempt HTTP/2 over ALPN with automatic HTTP/1.1 fallback when HTTP/2 is unsupported
+- Added priority service-tier traffic to the `premiumRequests` accounting on OpenAI and OpenAI Codex providers. Sending `serviceTier: "priority"` now increments `usage.premiumRequests` by 1 per request, matching the existing GitHub Copilot premium-request budget semantics so downstream consumers (e.g. the `omp stats` "Premium Reqs" card and `/usage`) reflect priority traffic alongside Copilot premium calls.
+
 ## [15.0.0] - 2026-05-13
 
 ### Added
@@ -12,6 +28,11 @@
 
 - Fixed OAuth credentials being silently disabled when two omp processes (or any two `AuthStorage` instances sharing a `agent.db`) race on token refresh. Anthropic rotates refresh tokens on every use, so the loser's `invalid_grant` response previously soft-deleted the row that the winner just rotated, forcing the user to `/login` again. `#tryOAuthCredential` now re-reads the row from disk before declaring a definitive failure: if the persisted `refresh` differs from the snapshot it tried, the peer-rotated credential is reloaded and the request retries against the fresh token instead of disabling the live row.
 - Closed a remaining race window in OAuth refresh-failure handling: between re-reading the credential row to check for peer rotation and the subsequent soft-delete, another process could still complete a refresh and rotate the row, leaving us to disable the freshly-rotated credential by `id`. The disable now runs as a single CAS update conditioned on the row's `data` still matching the snapshot we tried to refresh, and on `disabled_cause IS NULL`. If the CAS reports 0 rows changed (peer rotation, or row already disabled by a concurrent failure on the same snapshot), we reload from disk and retry instead of mutating the wrong row or emitting a spurious `credential_disabled` event.
+### Changed
+- Lowered the default steady-state stream idle timeout from 120s to 30s while preserving the existing environment overrides.
+
+### Fixed
+- Lazy built-in provider streams now enforce the shared idle watchdog and abort stalled provider requests, so session auto-retry can continue after transient network drops instead of remaining stuck. Caller aborts still terminate as aborted.
 
 ## [14.9.3] - 2026-05-10
 
@@ -2344,4 +2365,4 @@ _Dedicated to Peter's shoulder ([@steipete](https://twitter.com/steipete))_
 
 ## [0.9.4] - 2025-11-26
 
-Initial release with multi-provider LLM support.
+Initial release with multi-provider LLM support.

package/package.json

@@ -1,9 +1,9 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-ai",
-	"version": "15.0.0",
+	"version": "15.0.2",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
-	"homepage": "https://github.com/can1357/oh-my-pi",
+	"homepage": "https://omp.sh",
 	"author": "Can Boluk",
 	"contributors": [
 		"Mario Zechner"
@@ -46,8 +46,8 @@
 		"@aws-sdk/credential-provider-node": "^3.972.39",
 		"@bufbuild/protobuf": "^2.12.0",
 		"@google/genai": "^1.52.0",
-		"@oh-my-pi/pi-natives": "15.0.0",
-		"@oh-my-pi/pi-utils": "15.0.0",
+		"@oh-my-pi/pi-natives": "15.0.2",
+		"@oh-my-pi/pi-utils": "15.0.2",
 		"@sinclair/typebox": "^0.34.49",
 		"@smithy/node-http-handler": "^4.6.1",
 		"ajv": "^8.20.0",
@@ -58,10 +58,10 @@
 		"zod": "4.4.3"
 	},
 	"devDependencies": {
-		"@types/bun": "^1.3.13"
+		"@types/bun": "^1.3.14"
 	},
 	"engines": {
-		"bun": ">=1.3.7"
+		"bun": ">=1.3.14"
 	},
 	"files": [
 		"src",

package/src/index.ts

@@ -37,6 +37,7 @@ export * from "./usage/zai";
 export * from "./utils/anthropic-auth";
 export * from "./utils/discovery";
 export * from "./utils/event-stream";
+export * from "./utils/h2-fetch";
 export * from "./utils/overflow";
 export * from "./utils/retry";
 export * from "./utils/schema";

package/src/provider-models/ollama.ts

@@ -1,4 +1,4 @@
-import { abortableSleep } from "@oh-my-pi/pi-utils";
+import { fetchWithRetry } from "@oh-my-pi/pi-utils";
 import type { ModelManagerOptions } from "../model-manager";
 import { Effort } from "../model-thinking";
 import type { ThinkingConfig } from "../types";
@@ -19,16 +19,7 @@ type OllamaShowResponse = {
 	model_info?: Record<string, unknown>;
 };
 
-const MODEL_RETRY_DELAYS_MS = [2_000, 5_000, 10_000];
-
-async function fetchWithRetry(url: string, init: RequestInit): Promise<Response> {
-	for (let attempt = 0; attempt < MODEL_RETRY_DELAYS_MS.length; attempt++) {
-		const response = await fetch(url, init);
-		if (response.ok || response.status < 500) return response;
-		await abortableSleep(MODEL_RETRY_DELAYS_MS[attempt]!);
-	}
-	return fetch(url, init);
-}
+const OLLAMA_RETRY_DELAYS_MS = [2_000, 5_000, 10_000];
 
 function trimTrailingSlash(value: string): string {
 	return value.endsWith("/") ? value.slice(0, -1) : value;
@@ -109,6 +100,7 @@ export function ollamaCloudModelManagerOptions(
 			const response = await fetchWithRetry(`${baseUrl}/api/tags`, {
 				method: "GET",
 				headers: createCloudHeaders(apiKey),
+				defaultDelayMs: OLLAMA_RETRY_DELAYS_MS,
 			});
 			if (!response.ok) {
 				throw new Error(`HTTP ${response.status} from ${baseUrl}/api/tags`);

package/src/providers/anthropic.ts

@@ -1,5 +1,6 @@
 import * as nodeCrypto from "node:crypto";
 import * as fs from "node:fs";
+import { scheduler } from "node:timers/promises";
 import * as tls from "node:tls";
 import Anthropic, { type ClientOptions as AnthropicSdkClientOptions } from "@anthropic-ai/sdk";
 import type {
@@ -8,7 +9,14 @@ import type {
 	MessageParam,
 	RawMessageStreamEvent,
 } from "@anthropic-ai/sdk/resources/messages";
-import { $env, abortableSleep, isEnoent, readSseEvents } from "@oh-my-pi/pi-utils";
+import {
+	$env,
+	extractHttpStatusFromError,
+	isEnoent,
+	isRetryableError,
+	isUnexpectedSocketCloseMessage,
+	readSseEvents,
+} from "@oh-my-pi/pi-utils";
 import { hasOpus47ApiRestrictions, mapEffortToAnthropicAdaptiveEffort } from "../model-thinking";
 import { calculateCost } from "../models";
 import { getEnvApiKey, OUTPUT_FALLBACK_BUFFER } from "../stream";
@@ -17,6 +25,7 @@ import type {
 	AssistantMessage,
 	CacheRetention,
 	Context,
+	FetchImpl,
 	ImageContent,
 	Message,
 	Model,
@@ -48,12 +57,7 @@ import { getStreamFirstEventTimeoutMs, getStreamIdleTimeoutMs, iterateWithIdleTi
 import { parseJsonWithRepair, parseStreamingJson } from "../utils/json-parse";
 import { parseGitHubCopilotApiKey } from "../utils/oauth/github-copilot";
 import { notifyProviderResponse } from "../utils/provider-response";
-import {
-	extractHttpStatusFromError,
-	isCopilotRetryableError,
-	isRetryableError,
-	isUnexpectedSocketCloseMessage,
-} from "../utils/retry";
+import { isCopilotTransientModelError } from "../utils/retry";
 import { COMBINATOR_KEYS, NO_STRICT } from "../utils/schema";
 import { notifyRawSseEvent, wrapFetchForSseDebug } from "../utils/sse-debug";
 import {
@@ -538,6 +542,7 @@ export type AnthropicClientOptionsArgs = {
 	isOAuth?: boolean;
 	hasTools?: boolean;
 	onSseEvent?: AnthropicOptions["onSseEvent"];
+	fetch?: FetchImpl;
 };
 
 export type AnthropicClientOptionsResult = {
@@ -844,7 +849,7 @@ function isProviderRetryableStreamEnvelopeError(error: unknown): boolean {
 
 export function isProviderRetryableError(error: unknown, provider?: string): boolean {
 	if (!(error instanceof Error)) return false;
-	if (provider === "github-copilot" && isCopilotRetryableError(error)) return true;
+	if (provider === "github-copilot" && isCopilotTransientModelError(error)) return true;
 	const msg = error.message.toLowerCase();
 	if (
 		isUnexpectedSocketCloseMessage(msg) ||
@@ -962,6 +967,7 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 					isOAuth: options?.isOAuth,
 					hasTools: !!context.tools?.length,
 					onSseEvent: options?.onSseEvent,
+					fetch: options?.fetch,
 				});
 				client = created.client;
 				isOAuthToken = created.isOAuthToken;
@@ -1287,7 +1293,7 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 					}
 					providerRetryAttempt++;
 					const delayMs = PROVIDER_BASE_DELAY_MS * 2 ** (providerRetryAttempt - 1);
-					await abortableSleep(delayMs, options?.signal);
+					await scheduler.wait(delayMs, { signal: options?.signal });
 					output.content.length = 0;
 					output.responseId = undefined;
 					output.errorMessage = strictFallbackErrorMessage;
@@ -1402,7 +1408,12 @@ export function buildAnthropicClientOptions(args: AnthropicClientOptionsArgs): A
 	const baseUrl = resolveAnthropicBaseUrl(model, apiKey);
 	const foundryCustomHeaders = resolveAnthropicCustomHeaders(model);
 	const tlsFetchOptions = buildClaudeCodeTlsFetchOptions(model, baseUrl);
-	const debugFetch = onSseEvent ? wrapFetchForSseDebug(fetch, event => onSseEvent(event, model)) : undefined;
+	const baseFetch = args.fetch ?? fetch;
+	const debugFetch = onSseEvent
+		? wrapFetchForSseDebug(baseFetch, event => onSseEvent(event, model))
+		: args.fetch
+			? baseFetch
+			: undefined;
 	if (model.provider === "github-copilot") {
 		const copilotApiKey = parseGitHubCopilotApiKey(apiKey).accessToken;
 		const betaFeatures = [...extraBetas];

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

@@ -6,17 +6,15 @@ import type {
 	ResponseInput,
 } from "openai/resources/responses/responses";
 import { getEnvApiKey } from "../stream";
-import {
-	type Api,
-	type AssistantMessage,
-	type Context,
-	type Model,
-	type ServiceTier,
-	type StreamFunction,
-	type StreamOptions,
-	shouldSendServiceTier,
-	type Tool,
-	type ToolChoice,
+import type {
+	AssistantMessage,
+	Context,
+	Model,
+	ServiceTier,
+	StreamFunction,
+	StreamOptions,
+	Tool,
+	ToolChoice,
 } from "../types";
 import { normalizeSystemPrompts } from "../utils";
 import { createAbortSourceTracker } from "../utils/abort";
@@ -33,8 +31,11 @@ import { mapToOpenAIResponsesToolChoice } from "../utils/tool-choice";
 import { normalizeOpenAIResponsesPromptCacheKey, supportsDeveloperRole } from "./openai-responses";
 import {
 	appendResponsesToolResultMessages,
+	applyCommonResponsesSamplingParams,
+	applyResponsesReasoningParams,
 	convertResponsesAssistantMessage,
 	convertResponsesInputContent,
+	createInitialResponsesAssistantMessage,
 	normalizeResponsesToolCallIdForTransform,
 	processResponsesStream,
 } from "./openai-responses-shared";
@@ -101,23 +102,11 @@ export const streamAzureOpenAIResponses: StreamFunction<"azure-openai-responses"
 		let firstTokenTime: number | undefined;
 		const deploymentName = resolveDeploymentName(model, options);
 
-		const output: AssistantMessage = {
-			role: "assistant",
-			content: [],
-			api: "azure-openai-responses" as Api,
-			provider: model.provider,
-			model: model.id,
-			usage: {
-				input: 0,
-				output: 0,
-				cacheRead: 0,
-				cacheWrite: 0,
-				totalTokens: 0,
-				cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
-			},
-			stopReason: "stop",
-			timestamp: Date.now(),
-		};
+		const output: AssistantMessage = createInitialResponsesAssistantMessage(
+			"azure-openai-responses",
+			model.provider,
+			model.id,
+		);
 		let rawRequestDump: RawHttpRequestDump | undefined;
 		const abortTracker = createAbortSourceTracker(options?.signal);
 		const firstEventTimeoutAbortError = new Error(AZURE_OPENAI_RESPONSES_FIRST_EVENT_TIMEOUT_MESSAGE);
@@ -252,6 +241,7 @@ function createClient(model: Model<"azure-openai-responses">, apiKey: string, op
 
 	const { baseUrl, apiVersion } = resolveAzureConfig(model, options);
 
+	const baseFetch = options?.fetch ?? fetch;
 	return new AzureOpenAI({
 		apiKey,
 		apiVersion,
@@ -259,7 +249,9 @@ function createClient(model: Model<"azure-openai-responses">, apiKey: string, op
 		maxRetries: 5,
 		defaultHeaders: headers,
 		baseURL: baseUrl,
-		fetch: options?.onSseEvent ? wrapFetchForSseDebug(fetch, event => options.onSseEvent?.(event, model)) : fetch,
+		fetch: options?.onSseEvent
+			? wrapFetchForSseDebug(baseFetch, event => options.onSseEvent?.(event, model))
+			: baseFetch,
 	});
 }
 
@@ -279,31 +271,7 @@ function buildParams(
 		prompt_cache_key: normalizeOpenAIResponsesPromptCacheKey(options?.sessionId),
 	};
 
-	if (options?.maxTokens) {
-		params.max_output_tokens = options?.maxTokens;
-	}
-
-	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 (shouldSendServiceTier(options?.serviceTier, model.provider)) {
-		params.service_tier = options.serviceTier;
-	}
+	applyCommonResponsesSamplingParams(params, options, model.provider);
 
 	if (context.tools) {
 		params.tools = convertTools(context.tools);
@@ -312,36 +280,7 @@ function buildParams(
 		}
 	}
 
-	if (model.reasoning) {
-		// Always request encrypted reasoning content so reasoning items can be
-		// replayed in multi-turn conversations when store is false (items aren't
-		// persisted server-side, so we must include the full content).
-		// See: https://github.com/can1357/oh-my-pi/issues/41
-		params.include = ["reasoning.encrypted_content"];
-
-		if (options?.reasoning || options?.reasoningSummary !== undefined) {
-			const reasoningParams: NonNullable<typeof params.reasoning> = {
-				effort: options?.reasoning || "medium",
-			};
-			if (options?.reasoningSummary !== null) {
-				reasoningParams.summary = options?.reasoningSummary || "auto";
-			}
-			params.reasoning = reasoningParams;
-		} else {
-			if (model.name.toLowerCase().startsWith("gpt-5")) {
-				// Jesus Christ, see https://community.openai.com/t/need-reasoning-false-option-for-gpt-5/1351588/7
-				messages.push({
-					role: "developer",
-					content: [
-						{
-							type: "input_text",
-							text: "# Juice: 0 !important",
-						},
-					],
-				});
-			}
-		}
-	}
+	applyResponsesReasoningParams(params, model, options, messages);
 
 	return params;
 }

package/src/providers/gitlab-duo.ts

@@ -1,5 +1,5 @@
 import { ANTHROPIC_THINKING, mapAnthropicToolChoice } from "../stream";
-import type { Api, Context, Model, SimpleStreamOptions } from "../types";
+import type { Api, Context, FetchImpl, Model, SimpleStreamOptions } from "../types";
 import { AssistantMessageEventStream } from "../utils/event-stream";
 import type { OpenAICompletionsOptions } from "./openai-completions";
 import type { OpenAIResponsesOptions } from "./openai-responses";
@@ -172,13 +172,16 @@ interface DirectAccessToken {
 
 const directAccessCache = new Map<string, DirectAccessToken>();
 
-async function getDirectAccessToken(gitlabAccessToken: string): Promise<DirectAccessToken> {
+async function getDirectAccessToken(
+	gitlabAccessToken: string,
+	fetchImpl: FetchImpl = fetch,
+): Promise<DirectAccessToken> {
 	const cached = directAccessCache.get(gitlabAccessToken);
 	if (cached && cached.expiresAt > Date.now()) {
 		return cached;
 	}
 
-	const response = await fetch(`${GITLAB_COM_URL}/api/v4/ai/third_party_agents/direct_access`, {
+	const response = await fetchImpl(`${GITLAB_COM_URL}/api/v4/ai/third_party_agents/direct_access`, {
 		method: "POST",
 		headers: {
 			Authorization: `Bearer ${gitlabAccessToken}`,
@@ -240,7 +243,7 @@ export function streamGitLabDuo(
 				throw new Error(`Unsupported GitLab Duo model: ${model.id}`);
 			}
 
-			const directAccess = await getDirectAccessToken(options.apiKey);
+			const directAccess = await getDirectAccessToken(options.apiKey, options.fetch);
 			const headers = {
 				...directAccess.headers,
 				...options.headers,
@@ -278,6 +281,7 @@ export function streamGitLabDuo(
 								onPayload: options.onPayload,
 								onResponse: options.onResponse,
 								onSseEvent: options.onSseEvent,
+								fetch: options.fetch,
 								thinkingEnabled: Boolean(reasoningEffort) && model.reasoning,
 								thinkingBudgetTokens: reasoningEffort
 									? (options.thinkingBudgets?.[reasoningEffort] ?? ANTHROPIC_THINKING[reasoningEffort])
@@ -314,6 +318,7 @@ export function streamGitLabDuo(
 									onPayload: options.onPayload,
 									onResponse: options.onResponse,
 									onSseEvent: options.onSseEvent,
+									fetch: options.fetch,
 									reasoning: reasoningEffort,
 									toolChoice: options.toolChoice,
 								} satisfies OpenAIResponsesOptions,
@@ -345,6 +350,7 @@ export function streamGitLabDuo(
 									onPayload: options.onPayload,
 									onResponse: options.onResponse,
 									onSseEvent: options.onSseEvent,
+									fetch: options.fetch,
 									reasoning: reasoningEffort,
 									toolChoice: options.toolChoice,
 								} satisfies OpenAICompletionsOptions,