@eyepop.ai/eyepop 3.25.4 → 3.25.5

package/dist/eyepop.index.d.mts

@@ -64,6 +64,13 @@ interface Prediction extends StreamTime {
     source_url?: string;
     source_type?: string;
     system_timestamp?: number;
+    /**
+     * Compute units the frame's remote calls (a VLM ability's calls to vlm-server) used, in whole units. Absent
+     * when they used none, as for a local model.
+     */
+    compute_units?: number;
+    /** The same in milli compute units; from prediction version 4. */
+    milli_compute_units?: number;
     objects?: Array<PredictedObject>;
     classes?: Array<PredictedClass>;
     texts?: Array<PredictedText>;
@@ -1000,6 +1007,12 @@ interface AssetUuidSource {
 type Source = FileSource | StreamSource | PathSource | UrlSource | AssetUuidSource | MediaStreamSource;
 interface ResultStream extends AsyncIterable<Prediction> {
     cancel(): void;
+    /**
+     * What the stream's remote calls (a VLM ability's calls to vlm-server) used so far, in milli compute units:
+     * a running total while it runs, its total once it ended. Undefined until reported, and where a backend does
+     * not report it; a local model reports none. Informational: the called API bills its calls itself.
+     */
+    readonly milliComputeUnits?: number | undefined;
 }
 declare enum PopComponentType {
     FORWARD = "forward",
@@ -1325,9 +1338,15 @@ interface Options {
      */
     apiKey?: string | undefined;
     /**
-     * Bearer token accepted by the compute API.
+     * Bearer token accepted by the compute API: a login token, which the SDK exchanges at the
+     * compute API, or an already exchanged token, which it uses as is.
      */
     accessToken?: string | undefined;
+    /**
+     * The account a login token (`accessToken` or `oAuth2`) is exchanged for, sent to the compute API
+     * as `account_uuid`. Needed when the user has no single account to pick.
+     */
+    accountId?: string | undefined;
     /**
      * Temporary authentication token for client side execution.
      */
@@ -1425,6 +1444,7 @@ declare class Endpoint<T extends Endpoint<T>> {
     protected _options: Options;
     private _token;
     private _expire_token_time;
+    private _exchanging;
     protected _client: HttpClient | null;
     private _statusRetryHandlers;
     protected _limit: Semaphore;
@@ -1445,11 +1465,19 @@ declare class Endpoint<T extends Endpoint<T>> {
     connect(): Promise<T>;
     disconnect(wait?: boolean): Promise<void>;
     session(): Promise<Session>;
-    protected statusHandler401(_statusCode: number): void;
+    protected statusHandler401(_statusCode: number, failedToken?: string | null): void;
     protected fetchWithRetry(fetcher: () => Promise<Response>, retries?: number): Promise<Response>;
     protected reconnect(): Promise<T>;
     protected authorizationHeader(): Promise<string>;
+    private exchange;
+    private storeExchangedToken;
+    /**
+     * A token that already carries an account, or that the compute API refuses as an exchanged
+     * token ("API key or user login required"), is used as is. Any other refusal is raised.
+     */
+    private exchangeLoginToken;
     private currentAccessToken;
+    private authenticate;
 }
 
 /**
@@ -1635,9 +1663,15 @@ declare class InferAssetJob implements ResultStream {
     protected _inferRequest: InferRequest;
     protected _response: Promise<Prediction[]> | null;
     protected _runInfo: InferRunInfo | null;
+    protected _milliComputeUnits: number | undefined;
     protected _controller: AbortController;
     protected _requestLogger: Logger;
     [Symbol.asyncIterator](): AsyncIterator<Prediction>;
+    /**
+     * The milli compute units vlm-server billed for the inference, 0 on a cache hit; undefined before it answered
+     * or when it reported none.
+     */
+    get milliComputeUnits(): number | undefined;
     constructor(assetUrl: string, inferRequest: InferRequest, vlmFetch: VlmFetch, requestLogger: Logger);
     start(done: () => void): InferAssetJob;
     cancel(): void;
@@ -2111,8 +2145,13 @@ declare class PipelineJob implements ResultStream {
     readonly whenCancelled: Promise<void>;
     openSource: OpenSource | null;
     stopOnPipeline: (() => void) | null;
+    private _milliComputeUnits;
     constructor(source: Message, media: Media[], live: boolean, params: ProcessParams, mediaStream?: MediaStream | undefined);
     get cancelled(): boolean;
+    /** The source's milli compute units: the latest total of its `usage` messages and its `end`. */
+    get milliComputeUnits(): number | undefined;
+    /** Keep a usage's or an end's milliComputeUnits: each carries the source's total so far. */
+    noteMilliComputeUnits(message: Message): void;
     command(id: string): Message;
     predicted(prediction: Prediction): void;
     finish(error?: Error | null): void;

package/dist/eyepop.index.d.ts

@@ -64,6 +64,13 @@ interface Prediction extends StreamTime {
     source_url?: string;
     source_type?: string;
     system_timestamp?: number;
+    /**
+     * Compute units the frame's remote calls (a VLM ability's calls to vlm-server) used, in whole units. Absent
+     * when they used none, as for a local model.
+     */
+    compute_units?: number;
+    /** The same in milli compute units; from prediction version 4. */
+    milli_compute_units?: number;
     objects?: Array<PredictedObject>;
     classes?: Array<PredictedClass>;
     texts?: Array<PredictedText>;
@@ -1000,6 +1007,12 @@ interface AssetUuidSource {
 type Source = FileSource | StreamSource | PathSource | UrlSource | AssetUuidSource | MediaStreamSource;
 interface ResultStream extends AsyncIterable<Prediction> {
     cancel(): void;
+    /**
+     * What the stream's remote calls (a VLM ability's calls to vlm-server) used so far, in milli compute units:
+     * a running total while it runs, its total once it ended. Undefined until reported, and where a backend does
+     * not report it; a local model reports none. Informational: the called API bills its calls itself.
+     */
+    readonly milliComputeUnits?: number | undefined;
 }
 declare enum PopComponentType {
     FORWARD = "forward",
@@ -1325,9 +1338,15 @@ interface Options {
      */
     apiKey?: string | undefined;
     /**
-     * Bearer token accepted by the compute API.
+     * Bearer token accepted by the compute API: a login token, which the SDK exchanges at the
+     * compute API, or an already exchanged token, which it uses as is.
      */
     accessToken?: string | undefined;
+    /**
+     * The account a login token (`accessToken` or `oAuth2`) is exchanged for, sent to the compute API
+     * as `account_uuid`. Needed when the user has no single account to pick.
+     */
+    accountId?: string | undefined;
     /**
      * Temporary authentication token for client side execution.
      */
@@ -1425,6 +1444,7 @@ declare class Endpoint<T extends Endpoint<T>> {
     protected _options: Options;
     private _token;
     private _expire_token_time;
+    private _exchanging;
     protected _client: HttpClient | null;
     private _statusRetryHandlers;
     protected _limit: Semaphore;
@@ -1445,11 +1465,19 @@ declare class Endpoint<T extends Endpoint<T>> {
     connect(): Promise<T>;
     disconnect(wait?: boolean): Promise<void>;
     session(): Promise<Session>;
-    protected statusHandler401(_statusCode: number): void;
+    protected statusHandler401(_statusCode: number, failedToken?: string | null): void;
     protected fetchWithRetry(fetcher: () => Promise<Response>, retries?: number): Promise<Response>;
     protected reconnect(): Promise<T>;
     protected authorizationHeader(): Promise<string>;
+    private exchange;
+    private storeExchangedToken;
+    /**
+     * A token that already carries an account, or that the compute API refuses as an exchanged
+     * token ("API key or user login required"), is used as is. Any other refusal is raised.
+     */
+    private exchangeLoginToken;
     private currentAccessToken;
+    private authenticate;
 }
 
 /**
@@ -1635,9 +1663,15 @@ declare class InferAssetJob implements ResultStream {
     protected _inferRequest: InferRequest;
     protected _response: Promise<Prediction[]> | null;
     protected _runInfo: InferRunInfo | null;
+    protected _milliComputeUnits: number | undefined;
     protected _controller: AbortController;
     protected _requestLogger: Logger;
     [Symbol.asyncIterator](): AsyncIterator<Prediction>;
+    /**
+     * The milli compute units vlm-server billed for the inference, 0 on a cache hit; undefined before it answered
+     * or when it reported none.
+     */
+    get milliComputeUnits(): number | undefined;
     constructor(assetUrl: string, inferRequest: InferRequest, vlmFetch: VlmFetch, requestLogger: Logger);
     start(done: () => void): InferAssetJob;
     cancel(): void;
@@ -2111,8 +2145,13 @@ declare class PipelineJob implements ResultStream {
     readonly whenCancelled: Promise<void>;
     openSource: OpenSource | null;
     stopOnPipeline: (() => void) | null;
+    private _milliComputeUnits;
     constructor(source: Message, media: Media[], live: boolean, params: ProcessParams, mediaStream?: MediaStream | undefined);
     get cancelled(): boolean;
+    /** The source's milli compute units: the latest total of its `usage` messages and its `end`. */
+    get milliComputeUnits(): number | undefined;
+    /** Keep a usage's or an end's milliComputeUnits: each carries the source's total so far. */
+    noteMilliComputeUnits(message: Message): void;
     command(id: string): Message;
     predicted(prediction: Prediction): void;
     finish(error?: Error | null): void;

package/dist/eyepop.index.js

@@ -9044,14 +9044,25 @@ if ("document" in globalThis && "implementation" in globalThis.document) {
 }
 
 // endpoint.ts
+var ACCOUNT_CLAIM = "https://ident.eyepop.ai/account-uuid";
+var M2M_REFUSAL = "API key or user login required";
+function carriesAccount(token) {
+  try {
+    const payload = (token.split(".")[1] ?? "").replace(/-/g, "+").replace(/_/g, "/");
+    return Boolean(JSON.parse(atob(payload))?.[ACCOUNT_CLAIM]);
+  } catch {
+    return false;
+  }
+}
 var Endpoint = class {
   constructor(options) {
+    this._exchanging = null;
     this._options = options;
     this._token = null;
     this._expire_token_time = null;
     this._client = null;
     this._statusRetryHandlers = /* @__PURE__ */ new Map();
-    this._statusRetryHandlers.set(401, (statusCode) => this.statusHandler401(statusCode));
+    this._statusRetryHandlers.set(401, (statusCode, failedToken) => this.statusHandler401(statusCode, failedToken));
     this._limit = new Semaphore(this._options.jobQueueLength ?? 1024);
     this._state = "Idle" /* Idle */;
     this._message = null;
@@ -9067,7 +9078,7 @@ var Endpoint = class {
   }
   setStatusRetryHandlers(statusRetryHandlers = void 0) {
     this._statusRetryHandlers = new Map(statusRetryHandlers);
-    this._statusRetryHandlers.set(401, (statusCode) => this.statusHandler401(statusCode));
+    this._statusRetryHandlers.set(401, (statusCode, failedToken) => this.statusHandler401(statusCode, failedToken));
   }
   eyepopUrl() {
     if (this._options.eyepopUrl) {
@@ -9164,12 +9175,15 @@ var Endpoint = class {
       validUntil: this._expire_token_time * 1e3
     };
   }
-  statusHandler401(_statusCode) {
-    this._token = null;
-    this._expire_token_time = null;
+  statusHandler401(_statusCode, failedToken = this._token) {
+    if (this._token === failedToken) {
+      this._token = null;
+      this._expire_token_time = null;
+    }
   }
   async fetchWithRetry(fetcher, retries = 1) {
     while (true) {
+      const sentToken = this._token ?? void 0;
       try {
         const response = await fetcher();
         if (response.ok) {
@@ -9177,7 +9191,7 @@ var Endpoint = class {
         } else {
           const handler = this._statusRetryHandlers.get(response.status);
           if (handler != null) {
-            handler(response.status);
+            handler(response.status, sentToken);
             if (retries--) {
               this._requestLogger.info("received %d response, retry", response.status);
             } else {
@@ -9211,54 +9225,74 @@ var Endpoint = class {
     }
     return `Bearer ${await this.currentAccessToken()}`;
   }
+  async exchange(client, credential, accountUuid) {
+    const query = accountUuid ? `?account_uuid=${encodeURIComponent(accountUuid)}` : "";
+    return await client.fetch(`${this.eyepopUrl()}/v1/auth/authenticate${query}`, {
+      method: "POST",
+      headers: { Authorization: `Bearer ${credential}` }
+    });
+  }
+  async storeExchangedToken(response, now) {
+    if (response.status != 200) {
+      const message = await response.text();
+      return Promise.reject(`Unexpected status ${response.status}: ${message}`);
+    }
+    const token = await response.json();
+    this._token = token.access_token;
+    this._expire_token_time = now + token.expires_in - 60;
+  }
+  /**
+   * A token that already carries an account, or that the compute API refuses as an exchanged
+   * token ("API key or user login required"), is used as is. Any other refusal is raised.
+   */
+  async exchangeLoginToken(client, loginToken, validUntil, now) {
+    if (!carriesAccount(loginToken)) {
+      const response = await this.exchange(client, loginToken, this._options.accountId);
+      if (response.status != 401 || !(await response.clone().text()).includes(M2M_REFUSAL)) {
+        await this.storeExchangedToken(response, now);
+        return;
+      }
+    }
+    this._token = loginToken;
+    this._expire_token_time = validUntil;
+  }
   async currentAccessToken() {
+    if (!this._token || this._expire_token_time < Date.now() / 1e3) {
+      this._exchanging ?? (this._exchanging = this.authenticate().finally(() => this._exchanging = null));
+      await this._exchanging;
+    }
+    this._logger.debug("using access token, valid for at least %d seconds", this._expire_token_time - Date.now() / 1e3);
+    return this._token;
+  }
+  async authenticate() {
     const now = Date.now() / 1e3;
-    if (!this._token || this._expire_token_time < now) {
-      this.updateState("Authenticating" /* Authenticating */);
-      try {
-        const auth = resolveAuth(this._options);
-        if (!this._client) {
-          return Promise.reject("endpoint not connected");
-        } else if (auth?.isLocal !== void 0) {
-          this._token = "implicit";
-          this._expire_token_time = Number.MAX_VALUE;
-        } else if (auth?.session !== void 0) {
-          return Promise.reject("temporary access token expired");
-        } else if (auth?.oAuth2 !== void 0 && this._options.eyepopUrl) {
-          const session = await authenticateBrowserSession(auth.oAuth2, this._options);
-          this._token = session.accessToken;
-          this._expire_token_time = session.validUntil / 1e3;
-        } else if (accessTokenCredential(auth) !== void 0) {
-          this._token = accessTokenCredential(auth);
-          this._expire_token_time = Number.MAX_VALUE;
-        } else if (this._options.eyepopUrl) {
-          const credential = apiKeyCredential(auth);
-          if (credential === void 0) {
-            return Promise.reject("no valid auth option");
-          }
-          const headers = { Authorization: `Bearer ${credential}` };
-          const post_url = `${this.eyepopUrl()}/v1/auth/authenticate`;
-          const response = await this._client.fetch(post_url, {
-            method: "POST",
-            headers
-          });
-          if (response.status != 200) {
-            const message = await response.text();
-            return Promise.reject(`Unexpected status ${response.status}: ${message}`);
-          }
-          const data = await response.json();
-          const token = data;
-          this._token = token.access_token;
-          this._expire_token_time = now + token.expires_in - 60;
-        } else {
+    this.updateState("Authenticating" /* Authenticating */);
+    try {
+      const auth = resolveAuth(this._options);
+      if (!this._client) {
+        return Promise.reject("endpoint not connected");
+      } else if (auth?.isLocal !== void 0) {
+        this._token = "implicit";
+        this._expire_token_time = Number.MAX_VALUE;
+      } else if (auth?.session !== void 0) {
+        return Promise.reject("temporary access token expired");
+      } else if (auth?.oAuth2 !== void 0 && this._options.eyepopUrl) {
+        const login = await authenticateBrowserSession(auth.oAuth2, this._options);
+        await this.exchangeLoginToken(this._client, login.accessToken, login.validUntil / 1e3, now);
+      } else if (accessTokenCredential(auth) !== void 0) {
+        await this.exchangeLoginToken(this._client, accessTokenCredential(auth), Number.MAX_VALUE, now);
+      } else if (this._options.eyepopUrl) {
+        const credential = apiKeyCredential(auth);
+        if (credential === void 0) {
           return Promise.reject("no valid auth option");
         }
-      } finally {
-        this.updateState();
+        await this.storeExchangedToken(await this.exchange(this._client, credential), now);
+      } else {
+        return Promise.reject("no valid auth option");
       }
+    } finally {
+      this.updateState();
     }
-    this._logger.debug("using access token, valid for at least %d seconds", this._expire_token_time - now);
-    return this._token;
   }
 };
 
@@ -11808,10 +11842,22 @@ var PipelineJob = class {
     this.openSource = null;
     // Stops the job's source on the pipeline, once it was sent.
     this.stopOnPipeline = null;
+    this._milliComputeUnits = void 0;
   }
   get cancelled() {
     return this._cancelled;
   }
+  /** The source's milli compute units: the latest total of its `usage` messages and its `end`. */
+  get milliComputeUnits() {
+    return this._milliComputeUnits;
+  }
+  /** Keep a usage's or an end's milliComputeUnits: each carries the source's total so far. */
+  noteMilliComputeUnits(message) {
+    const value = message["milliComputeUnits"];
+    if (Number.isInteger(value) && value >= 0) {
+      this._milliComputeUnits = value;
+    }
+  }
   command(id) {
     const command = { type: "setSource", id, ...this.source };
     if (this.params.componentParams !== void 0) {
@@ -12389,7 +12435,11 @@ var PipelineBackend = class {
         case "prediction":
           job.predicted(message["prediction"]);
           break;
+        case "usage":
+          job.noteMilliComputeUnits(message);
+          break;
         case "end":
+          job.noteMilliComputeUnits(message);
           if (message["reason"] === "eos") {
             return null;
           }
@@ -12953,6 +13003,7 @@ var InferAssetJob = class {
   constructor(assetUrl, inferRequest, vlmFetch, requestLogger) {
     this._response = null;
     this._runInfo = null;
+    this._milliComputeUnits = void 0;
     this._controller = new AbortController();
     this._vlmFetch = vlmFetch;
     this._assetUrl = assetUrl;
@@ -12989,6 +13040,13 @@ var InferAssetJob = class {
     }
     return iterator();
   }
+  /**
+   * The milli compute units vlm-server billed for the inference, 0 on a cache hit; undefined before it answered
+   * or when it reported none.
+   */
+  get milliComputeUnits() {
+    return this._milliComputeUnits;
+  }
   start(done) {
     this._response = new Promise(async (resolve, reject) => {
       try {
@@ -13013,6 +13071,8 @@ var InferAssetJob = class {
           const runInfo = response["run_info"];
           if (predictions) {
             this._runInfo = runInfo;
+            const used = response["milli_compute_units"] ?? response["compute_units"];
+            this._milliComputeUnits = typeof used === "number" ? used : void 0;
             resolve(predictions);
             break;
           } else {