@oh-my-pi/pi-ai 15.7.3 → 15.7.6

package/CHANGELOG.md

@@ -2,9 +2,23 @@
 
 ## [Unreleased]
 
+## [15.7.5] - 2026-06-01
+
+### Added
+
+- Added Anthropic task budget support, forwarding `taskBudget` as `output_config.task_budget` with the required `task-budgets-2026-03-13` beta header and accepting Anthropic gateway requests that send `output_config.task_budget`.
+
+### Fixed
+
+- Fixed OpenAI-family first-event timeouts so `PI_OPENAI_STREAM_IDLE_TIMEOUT_MS` cannot be undercut by a lower generic `PI_STREAM_FIRST_EVENT_TIMEOUT_MS` while local OpenAI-compatible servers are still processing large prompts. `PI_OPENAI_STREAM_FIRST_EVENT_TIMEOUT_MS` is now available for an explicit OpenAI-specific first-event override. ([#1603](https://github.com/can1357/oh-my-pi/issues/1603))
+
+## [15.7.4] - 2026-05-31
+
 ### Fixed
 
 - Fixed Anthropic stream idle-timeout retries after the provider stream has already begun.
+- Fixed Xiaomi MiMo `/login` rejecting token-plan (`tp-`) keys with `401 Invalid API Key`. The validation request was still sending the legacy Anthropic `x-api-key` header against the OpenAI-compatible `/v1/chat/completions` endpoint; switched to `Authorization: Bearer`, matching the runtime path. ([#1580](https://github.com/can1357/oh-my-pi/issues/1580))
+- Fixed OpenAI-compatible tool-call replay to send empty assistant content instead of `null`, avoiding strict custom backends that crash with `str`/`NoneType` concatenation after subagent tool results. ([#1585](https://github.com/can1357/oh-my-pi/issues/1585))
 
 ## [15.7.3] - 2026-05-31
 

package/dist/types/auth-gateway/types.d.ts

@@ -1,5 +1,5 @@
 import type { Effort } from "../model-thinking";
-import type { AssistantMessage, AssistantMessageEventStream, CacheRetention, Context, ServiceTier } from "../types";
+import type { AssistantMessage, AssistantMessageEventStream, CacheRetention, Context, ServiceTier, TokenTaskBudget } from "../types";
 /**
  * Wire types for the omp auth-gateway.
  *
@@ -54,6 +54,8 @@ export interface AuthGatewayParsedRequestOptions {
     thinkingBudgets?: Partial<Record<Effort, number>>;
     /** Suppress the provider's reasoning summary stream. */
     hideThinkingSummary?: boolean;
+    /** Anthropic `output_config.task_budget` advisory loop budget. */
+    taskBudget?: TokenTaskBudget;
     /** OpenAI service tier (auto|default|flex|scale|priority). */
     serviceTier?: ServiceTier;
     /** Cache retention hint derived from inbound `cache_control` markers. */

package/dist/types/providers/anthropic-messages-server-schema.d.ts

@@ -429,6 +429,21 @@ export declare const anthropicMessagesRequestSchema: z.ZodObject<{
         budget_tokens: z.ZodOptional<z.ZodNumber>;
         display: z.ZodOptional<z.ZodUnknown>;
     }, z.core.$strip>], "type">>;
+    output_config: z.ZodOptional<z.ZodObject<{
+        effort: z.ZodOptional<z.ZodEnum<{
+            high: "high";
+            low: "low";
+            max: "max";
+            medium: "medium";
+            xhigh: "xhigh";
+        }>>;
+        task_budget: z.ZodOptional<z.ZodObject<{
+            type: z.ZodLiteral<"tokens">;
+            total: z.ZodNumber;
+            remaining: z.ZodOptional<z.ZodNumber>;
+        }, z.core.$strip>>;
+        format: z.ZodOptional<z.ZodUnknown>;
+    }, z.core.$strip>>;
     metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
     container: z.ZodOptional<z.ZodUnknown>;
     context_management: z.ZodOptional<z.ZodUnknown>;

package/dist/types/types.d.ts

@@ -55,6 +55,11 @@ import type { Effort } from "./model-thinking";
 export type ThinkingBudgets = {
     [key in Effort]?: number;
 };
+export interface TokenTaskBudget {
+    type: "tokens";
+    total: number;
+    remaining?: number;
+}
 export type MessageAttribution = "user" | "agent";
 export type ToolChoice = "auto" | "none" | "any" | "required" | {
     type: "function";
@@ -184,6 +189,11 @@ export interface StreamOptions {
      * For example, Anthropic uses `user_id` for abuse tracking and rate limiting.
      */
     metadata?: Record<string, unknown>;
+    /**
+     * Advisory token budget for a full agentic loop. Anthropic encodes this as
+     * `output_config.task_budget` with the `task-budgets-2026-03-13` beta header.
+     */
+    taskBudget?: TokenTaskBudget;
     /**
      * Optional session identifier for providers that support session-based
      * routing, request affinity, or transport reuse. Providers may also use this
@@ -228,6 +238,11 @@ export interface StreamOptions {
      * `0` to disable both layers for this request. After the first semantic
      * event arrives, `streamIdleTimeoutMs` governs inter-event stalls. Falls
      * back to `PI_STREAM_FIRST_EVENT_TIMEOUT_MS` and then to a 100s default.
+     * OpenAI-family transports additionally honor
+     * `PI_OPENAI_STREAM_FIRST_EVENT_TIMEOUT_MS` as the most-specific override and
+     * floor the first-event budget at the resolved idle (per-call
+     * `streamIdleTimeoutMs` or `PI_OPENAI_STREAM_IDLE_TIMEOUT_MS`) so slow local
+     * OpenAI-compatible servers are not undercut during prompt processing.
      *
      * Iterator-level honored by: every built-in provider (via the lazy-stream
      * forwarder in `register-builtins`). SDK-request honored by:

package/dist/types/utils/idle-iterator.d.ts

@@ -32,6 +32,20 @@ export declare function getOpenAIStreamIdleTimeoutMs(fallbackMs?: number): numbe
  * env overrides still trump the fallback.
  */
 export declare function getStreamFirstEventTimeoutMs(idleTimeoutMs?: number, fallbackMs?: number): number | undefined;
+/**
+ * Returns the first-event timeout used for OpenAI-family streaming transports.
+ *
+ * Precedence: explicit `PI_OPENAI_STREAM_FIRST_EVENT_TIMEOUT_MS` (including a
+ * `"0"` disable) wins outright. Otherwise the resolved idle (caller-supplied
+ * `idleTimeoutMs` — which itself already encompasses per-call
+ * `streamIdleTimeoutMs` or `PI_OPENAI_STREAM_IDLE_TIMEOUT_MS` resolved
+ * upstream) floors the first-event budget so slow local OpenAI-compatible
+ * servers are not undercut by a shorter `PI_STREAM_FIRST_EVENT_TIMEOUT_MS`
+ * or the global default during prompt processing.
+ *
+ * Returns `undefined` when an explicit env knob disables the watchdog.
+ */
+export declare function getOpenAIStreamFirstEventTimeoutMs(idleTimeoutMs?: number, fallbackMs?: number): number | undefined;
 export interface IdleTimeoutIteratorOptions {
     idleTimeoutMs?: number;
     firstItemTimeoutMs?: number;

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-ai",
-	"version": "15.7.3",
+	"version": "15.7.6",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"homepage": "https://omp.sh",
 	"author": "Can Boluk",
@@ -40,7 +40,7 @@
 	"dependencies": {
 		"@anthropic-ai/sdk": "^0.99.0",
 		"@bufbuild/protobuf": "^2.12.0",
-		"@oh-my-pi/pi-utils": "15.7.3",
+		"@oh-my-pi/pi-utils": "15.7.6",
 		"openai": "^6.39.0",
 		"partial-json": "^0.1.7",
 		"zod": "4.4.3"

package/src/auth-gateway/server.ts

@@ -141,6 +141,7 @@ function buildStreamOptions(parsed: ParsedFormatRequest, api: Api, signal: Abort
 	if (options.reasoning !== undefined) opts.reasoning = options.reasoning;
 	if (options.disableReasoning !== undefined) opts.disableReasoning = options.disableReasoning;
 	if (options.hideThinkingSummary !== undefined) opts.hideThinkingSummary = options.hideThinkingSummary;
+	if (options.taskBudget !== undefined) opts.taskBudget = options.taskBudget;
 	if (options.serviceTier !== undefined) opts.serviceTier = options.serviceTier;
 	if (options.cacheRetention !== undefined) opts.cacheRetention = options.cacheRetention;
 	// Client-supplied `prompt_cache_key` wins; otherwise derive a stable

package/src/auth-gateway/types.ts

@@ -1,5 +1,12 @@
 import type { Effort } from "../model-thinking";
-import type { AssistantMessage, AssistantMessageEventStream, CacheRetention, Context, ServiceTier } from "../types";
+import type {
+	AssistantMessage,
+	AssistantMessageEventStream,
+	CacheRetention,
+	Context,
+	ServiceTier,
+	TokenTaskBudget,
+} from "../types";
 
 /**
  * Wire types for the omp auth-gateway.
@@ -61,6 +68,8 @@ export interface AuthGatewayParsedRequestOptions {
 	thinkingBudgets?: Partial<Record<Effort, number>>;
 	/** Suppress the provider's reasoning summary stream. */
 	hideThinkingSummary?: boolean;
+	/** Anthropic `output_config.task_budget` advisory loop budget. */
+	taskBudget?: TokenTaskBudget;
 
 	// ── Service / routing ─────────────────────────────────────────────────
 	/** OpenAI service tier (auto|default|flex|scale|priority). */