aisubs 0.3.3 → 0.3.5

package/CHANGELOG.md

@@ -1,5 +1,17 @@
 # Changelog
 
+## 0.3.5 - 2026-09-14
+
+- Move ChatGPT Responses instructions into the developer input prefix so
+  stable subscription prompts participate in prompt caching.
+
+## 0.3.4 - 2026-09-13
+
+- Refresh account model catalogs on demand and update ChatGPT compatibility so
+  newly available subscription models appear without stale cached listings.
+- Preserve prompt-cache routing and cache read/write usage when translating
+  between OpenAI, Anthropic, and Responses-compatible protocols.
+
 ## 0.3.3 - 2026-09-02
 
 - Show remaining usage percentages in account meter values and progress bars,

package/dist/auth.d.ts

@@ -19,7 +19,7 @@ export interface SubscriptionAccount {
     fetch(input: string | URL | Request, init?: RequestInit): Promise<Response>;
     proxy(path: string, init?: RequestInit): Promise<Response>;
     getUsage(signal?: AbortSignal): Promise<ProviderUsage | null>;
-    getModels(signal?: AbortSignal): Promise<ProviderModels | null>;
+    getModels(signal?: AbortSignal, force?: boolean): Promise<ProviderModels | null>;
     credentialSummary(): Promise<CredentialSummary>;
     details(signal?: AbortSignal): Promise<SubscriptionAccountDetails>;
 }
@@ -69,7 +69,7 @@ export declare class SubscriptionAuth {
     /** Build an authorized direct-provider request for transports such as WebSocket. */
     authorizeProxyRequest(provider: ProviderId, account: string, path: string, init?: RequestInit): Promise<Request>;
     getUsage(provider: ProviderId, account?: string, callerSignal?: AbortSignal): Promise<ProviderUsage | null>;
-    getModels(provider: ProviderId, account?: string, callerSignal?: AbortSignal): Promise<ProviderModels | null>;
+    getModels(provider: ProviderId, account?: string, callerSignal?: AbortSignal, force?: boolean): Promise<ProviderModels | null>;
     account(provider: ProviderId, account: string): SubscriptionAccount;
 }
 export declare function createSubscriptionAuth(options: {

package/dist/auth.js

@@ -459,12 +459,17 @@ export class SubscriptionAuth {
             return data ? { provider, accountKey, asOf: Date.now(), ...data } : null;
         });
     }
-    async getModels(provider, account = DEFAULT_ACCOUNT, callerSignal) {
+    async getModels(provider, account = DEFAULT_ACCOUNT, callerSignal, force = false) {
         const adapter = this.adapter(provider);
         if (!adapter.getModels)
             return null;
         const accountKey = normalizeAccountKey(account);
         const scope = credentialKey(provider, accountKey);
+        if (force) {
+            this.modelsCache.delete(scope);
+            this.modelsInflight.delete(scope);
+            this.metadataGenerations.set(scope, (this.metadataGenerations.get(scope) ?? 0) + 1);
+        }
         return this.cachedMetadata(this.modelsCache, this.modelsInflight, scope, this.modelsCacheTtlMs, callerSignal, async () => {
             const credential = await this.credential(provider, accountKey);
             const timeout = AbortSignal.timeout(this.refreshTimeoutMs);
@@ -490,7 +495,7 @@ export class SubscriptionAuth {
             fetch: (input, init) => this.fetch(provider, input, init, accountKey),
             proxy: (path, init) => this.proxy(provider, accountKey, path, init),
             getUsage: (signal) => this.getUsage(provider, accountKey, signal),
-            getModels: (signal) => this.getModels(provider, accountKey, signal),
+            getModels: (signal, force) => this.getModels(provider, accountKey, signal, force),
             credentialSummary: () => this.credentialSummary(provider, accountKey),
             details: (signal) => this.details(provider, accountKey, signal),
         };

package/dist/compatibility.js

@@ -150,6 +150,7 @@ function parseChat(body) {
         responseFormat: raw.response_format,
         metadata: raw.metadata,
         user: stringValue(raw.user),
+        promptCacheKey: stringValue(raw.prompt_cache_key),
     };
 }
 function responseTools(value) {
@@ -243,6 +244,7 @@ function parseResponses(body) {
         responseFormat,
         metadata: raw.metadata,
         user: stringValue(raw.user),
+        promptCacheKey: stringValue(raw.prompt_cache_key),
     };
 }
 function anthropicParts(value) {
@@ -472,6 +474,7 @@ function toChat(request) {
         ...(request.responseFormat != null ? { response_format: request.responseFormat } : {}),
         ...(request.metadata != null ? { metadata: request.metadata } : {}),
         ...(request.user ? { user: request.user } : {}),
+        ...(request.promptCacheKey ? { prompt_cache_key: request.promptCacheKey } : {}),
     };
 }
 function responseContent(part, role) {
@@ -557,6 +560,7 @@ function toResponses(request) {
         ...(format != null ? { text: { format } } : {}),
         ...(request.metadata != null ? { metadata: request.metadata } : {}),
         ...(request.user ? { user: request.user } : {}),
+        ...(request.promptCacheKey ? { prompt_cache_key: request.promptCacheKey } : {}),
     };
 }
 function anthropicContent(message) {
@@ -703,7 +707,7 @@ function toGoogle(request) {
         },
     };
 }
-function usage(input, output, cached, reasoning) {
+function usage(input, output, cached, reasoning, cacheWrite) {
     if (input == null && output == null)
         return undefined;
     return {
@@ -711,6 +715,7 @@ function usage(input, output, cached, reasoning) {
         output: output ?? 0,
         total: (input ?? 0) + (output ?? 0),
         ...(cached != null ? { cached } : {}),
+        ...(cacheWrite != null ? { cacheWrite } : {}),
         ...(reasoning != null ? { reasoning } : {}),
     };
 }
@@ -735,7 +740,7 @@ function parseChatResult(raw, model) {
         finishReason: finish === "length" || finish === "tool_calls" || finish === "content_filter"
             ? finish
             : "stop",
-        usage: usage(numberValue(details?.prompt_tokens), numberValue(details?.completion_tokens), numberValue(promptDetails?.cached_tokens), numberValue(completionDetails?.reasoning_tokens)),
+        usage: usage(numberValue(details?.prompt_tokens), numberValue(details?.completion_tokens), numberValue(promptDetails?.cached_tokens), numberValue(completionDetails?.reasoning_tokens), numberValue(promptDetails?.cache_write_tokens)),
     };
 }
 function parseResponsesResult(raw, model) {
@@ -785,7 +790,7 @@ function parseResponsesResult(raw, model) {
                 : raw.status === "failed"
                     ? "error"
                     : "stop",
-        usage: usage(numberValue(details?.input_tokens), numberValue(details?.output_tokens), numberValue(inputDetails?.cached_tokens), numberValue(outputDetails?.reasoning_tokens)),
+        usage: usage(numberValue(details?.input_tokens), numberValue(details?.output_tokens), numberValue(inputDetails?.cached_tokens), numberValue(outputDetails?.reasoning_tokens), numberValue(inputDetails?.cache_write_tokens)),
     };
 }
 function parseAnthropicResult(raw, model) {
@@ -805,7 +810,11 @@ function parseAnthropicResult(raw, model) {
                 : stop === "refusal"
                     ? "content_filter"
                     : "stop",
-        usage: usage(numberValue(rawUsage?.input_tokens), numberValue(rawUsage?.output_tokens), numberValue(rawUsage?.cache_read_input_tokens)),
+        usage: usage(rawUsage?.input_tokens != null
+            ? (numberValue(rawUsage.input_tokens) ?? 0) +
+                (numberValue(rawUsage.cache_read_input_tokens) ?? 0) +
+                (numberValue(rawUsage.cache_creation_input_tokens) ?? 0)
+            : undefined, numberValue(rawUsage?.output_tokens), numberValue(rawUsage?.cache_read_input_tokens), undefined, numberValue(rawUsage?.cache_creation_input_tokens)),
     };
 }
 function parseGoogleResult(raw, model) {
@@ -876,8 +885,15 @@ function resultToChat(result) {
                     prompt_tokens: result.usage.input,
                     completion_tokens: result.usage.output,
                     total_tokens: result.usage.total,
-                    ...(result.usage.cached != null
-                        ? { prompt_tokens_details: { cached_tokens: result.usage.cached } }
+                    ...(result.usage.cached != null || result.usage.cacheWrite != null
+                        ? {
+                            prompt_tokens_details: {
+                                ...(result.usage.cached != null ? { cached_tokens: result.usage.cached } : {}),
+                                ...(result.usage.cacheWrite != null
+                                    ? { cache_write_tokens: result.usage.cacheWrite }
+                                    : {}),
+                            },
+                        }
                         : {}),
                     ...(result.usage.reasoning != null
                         ? { completion_tokens_details: { reasoning_tokens: result.usage.reasoning } }
@@ -935,7 +951,16 @@ function resultToResponses(result) {
                     input_tokens: result.usage.input,
                     output_tokens: result.usage.output,
                     total_tokens: result.usage.total,
-                    input_tokens_details: { cached_tokens: result.usage.cached ?? 0 },
+                    ...(result.usage.cached != null || result.usage.cacheWrite != null
+                        ? {
+                            input_tokens_details: {
+                                ...(result.usage.cached != null ? { cached_tokens: result.usage.cached } : {}),
+                                ...(result.usage.cacheWrite != null
+                                    ? { cache_write_tokens: result.usage.cacheWrite }
+                                    : {}),
+                            },
+                        }
+                        : {}),
                     output_tokens_details: { reasoning_tokens: result.usage.reasoning ?? 0 },
                 },
             }
@@ -965,11 +990,14 @@ function resultToAnthropic(result) {
         ...(result.usage
             ? {
                 usage: {
-                    input_tokens: result.usage.input,
+                    input_tokens: Math.max(0, result.usage.input - (result.usage.cached ?? 0) - (result.usage.cacheWrite ?? 0)),
                     output_tokens: result.usage.output,
                     ...(result.usage.cached != null
                         ? { cache_read_input_tokens: result.usage.cached }
                         : {}),
+                    ...(result.usage.cacheWrite != null
+                        ? { cache_creation_input_tokens: result.usage.cacheWrite }
+                        : {}),
                 },
             }
             : {}),
@@ -1269,7 +1297,7 @@ function streamAnthropic(result) {
             ...message,
             content: [],
             stop_reason: null,
-            usage: { input_tokens: result.usage?.input ?? 0, output_tokens: 0 },
+            ...(message.usage ? { usage: { ...message.usage, output_tokens: 0 } } : {}),
         },
     });
     for (const [index, block] of message.content.entries()) {

package/dist/providers/chatgpt.js

@@ -109,6 +109,25 @@ async function normalizeChatGptRequest(request) {
     if (!isRecord(raw))
         return request;
     const body = { ...raw };
+    // The Codex endpoint places the implicit cache boundary after input
+    // messages. Move stable top-level instructions into that prefix so they
+    // participate in prompt caching for subscription requests.
+    if (typeof body.instructions === "string" && body.instructions.length > 0) {
+        const input = Array.isArray(body.input)
+            ? body.input
+            : typeof body.input === "string"
+                ? [{ type: "message", role: "user", content: body.input }]
+                : [];
+        body.input = [
+            {
+                type: "message",
+                role: "developer",
+                content: [{ type: "input_text", text: body.instructions }],
+            },
+            ...input,
+        ];
+        delete body.instructions;
+    }
     delete body.prompt_cache_options;
     delete body.prompt_cache_retention;
     const stripBreakpoints = (value) => Array.isArray(value)
@@ -126,7 +145,7 @@ async function normalizeChatGptRequest(request) {
 }
 export function chatGptProvider(options = {}) {
     const clientId = options.clientId ?? DEFAULT_CLIENT_ID;
-    const compatibilityVersion = options.compatibilityVersion ?? "0.144.2";
+    const compatibilityVersion = options.compatibilityVersion ?? "0.154.0";
     const fetcher = options.fetch ?? globalThis.fetch;
     async function startDeviceLogin(signal) {
         const response = await fetcher(DEVICE_CODE_URL, {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "aisubs",
-  "version": "0.3.3",
+  "version": "0.3.5",
   "description": "Connect AI provider accounts and use those subscriptions from any local tool or as api.",
   "keywords": [
     "ai",