@canarycoders/ai 0.5.0 → 0.6.0

package/README.md

@@ -161,6 +161,10 @@ await oa.chat.completions.create({
 | `defaultTag` | — | attached to requests for usage attribution |
 | `poll` | — | default poll timing for queued ops |
 
+## Agent skill
+
+`skills/canary-ai/` ships a Codex/Claude Code Agent Skill that drives the gateway (research, X search, chat, image generation, model discovery) from a single Bun script, no MCP server needed. See [docs.ai.canarycoders.es/agent-skill](https://docs.ai.canarycoders.es/agent-skill) for install/update/uninstall steps. It's a plain repo folder, not part of the published npm package.
+
 ## Keeping types in sync
 
 The gateway API is the source of truth. Regenerate types from its OpenAPI spec and diff them against the hand-written ones:

package/dist/index.d.cts

@@ -6,15 +6,18 @@ export { b as RealtimeKind, c as RealtimeTool } from './realtime-BWDkXcj9.cjs';
 type FetchLike = typeof globalThis.fetch;
 
 /** Every provider the gateway can route to. */
-type ProviderType = "gemini" | "vertex" | "openai" | "anthropic" | "xai" | "lmstudio" | "ollama" | "perplexity" | "elevenlabs" | "mlxaudio" | "vision" | "gcvision";
+type ProviderType = "gemini" | "vertex" | "openai" | "anthropic" | "xai" | "lmstudio" | "ollama" | "perplexity" | "elevenlabs" | "mlxaudio" | "vision" | "gcvision" | "zhipu" | "moonshot";
 /** Providers that serve text chat / completion. */
-type ChatProvider = "gemini" | "vertex" | "openai" | "anthropic" | "xai" | "lmstudio" | "ollama" | "perplexity";
+type ChatProvider = "gemini" | "vertex" | "openai" | "anthropic" | "xai" | "lmstudio" | "ollama" | "perplexity" | "zhipu" | "moonshot";
 type ModelCapability = "chat" | "vision" | "image-generation" | "video-generation" | "text-to-speech" | "speech-to-text" | "image-recognition" | "face-detection" | "web-detection" | "conversation" | "sound-effect" | "music-generation" | "dialogue" | "realtime-voice" | "realtime-translate" | "realtime-transcribe" | "embeddings" | "reasoning";
 interface TokenUsage {
     inputTokens: number;
     outputTokens: number;
     totalTokens: number;
+    /** Cache-read tokens (subset of inputTokens), billed at the cached-input rate. */
     cachedTokens?: number;
+    /** Anthropic only: cache-write tokens (billed at 1.25x input). */
+    cacheCreationTokens?: number;
     reasoningTokens?: number;
 }
 interface ModelInfo {
@@ -297,6 +300,15 @@ interface CompleteParams {
     tools?: ToolDefinition[];
     toolChoice?: ToolChoice;
     thinkingMode?: ThinkingMode;
+    /**
+     * Anthropic only: opt into prompt caching. The gateway places cache_control
+     * breakpoints on the system prompt and the last message (5-minute TTL).
+     */
+    promptCaching?: boolean;
+    /** OpenAI only: stable cache-routing key (per conversation/workload) to raise automatic cache hits. */
+    promptCacheKey?: string;
+    /** OpenAI only: stable end-user identifier, also used for cache routing. */
+    safetyIdentifier?: string;
     webSearch?: WebSearchOptions;
     cache?: {
         enabled?: boolean;

package/dist/index.d.ts

@@ -6,15 +6,18 @@ export { b as RealtimeKind, c as RealtimeTool } from './realtime-BWDkXcj9.js';
 type FetchLike = typeof globalThis.fetch;
 
 /** Every provider the gateway can route to. */
-type ProviderType = "gemini" | "vertex" | "openai" | "anthropic" | "xai" | "lmstudio" | "ollama" | "perplexity" | "elevenlabs" | "mlxaudio" | "vision" | "gcvision";
+type ProviderType = "gemini" | "vertex" | "openai" | "anthropic" | "xai" | "lmstudio" | "ollama" | "perplexity" | "elevenlabs" | "mlxaudio" | "vision" | "gcvision" | "zhipu" | "moonshot";
 /** Providers that serve text chat / completion. */
-type ChatProvider = "gemini" | "vertex" | "openai" | "anthropic" | "xai" | "lmstudio" | "ollama" | "perplexity";
+type ChatProvider = "gemini" | "vertex" | "openai" | "anthropic" | "xai" | "lmstudio" | "ollama" | "perplexity" | "zhipu" | "moonshot";
 type ModelCapability = "chat" | "vision" | "image-generation" | "video-generation" | "text-to-speech" | "speech-to-text" | "image-recognition" | "face-detection" | "web-detection" | "conversation" | "sound-effect" | "music-generation" | "dialogue" | "realtime-voice" | "realtime-translate" | "realtime-transcribe" | "embeddings" | "reasoning";
 interface TokenUsage {
     inputTokens: number;
     outputTokens: number;
     totalTokens: number;
+    /** Cache-read tokens (subset of inputTokens), billed at the cached-input rate. */
     cachedTokens?: number;
+    /** Anthropic only: cache-write tokens (billed at 1.25x input). */
+    cacheCreationTokens?: number;
     reasoningTokens?: number;
 }
 interface ModelInfo {
@@ -297,6 +300,15 @@ interface CompleteParams {
     tools?: ToolDefinition[];
     toolChoice?: ToolChoice;
     thinkingMode?: ThinkingMode;
+    /**
+     * Anthropic only: opt into prompt caching. The gateway places cache_control
+     * breakpoints on the system prompt and the last message (5-minute TTL).
+     */
+    promptCaching?: boolean;
+    /** OpenAI only: stable cache-routing key (per conversation/workload) to raise automatic cache hits. */
+    promptCacheKey?: string;
+    /** OpenAI only: stable end-user identifier, also used for cache routing. */
+    safetyIdentifier?: string;
     webSearch?: WebSearchOptions;
     cache?: {
         enabled?: boolean;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@canarycoders/ai",
-  "version": "0.5.0",
+  "version": "0.6.0",
   "description": "Official TypeScript SDK for the CanaryCoders AI multi-provider gateway",
   "license": "MIT",
   "author": "CanaryCoders",