skytells 1.0.2 → 1.0.4

package/dist/client.cjs

@@ -1,59 +1,1175 @@
-const { HTTP } = require("./http.js");
-const { ENDPOINTS } = require("./endpoints.js");
+/**
+ * Skytells prediction client: models catalog, create/poll/wait predictions, queue batching, and
+ * optional per-request inference compatibility guard (second argument `{ compatibilityCheck: true }` on `predict` / `create`, fourth on `run`).
+ *
+ * **Memory / lifecycle (for long‑lived servers):**
+ * - Sub-APIs (`chat`, `embeddings`, …) are created lazily on first access.
+ * - When a call passes `{ compatibilityCheck: true }`, model metadata is cached per slug with TTL
+ *   {@link PREFETCHED_MODEL_CACHE_TTL_MS} and a cap from {@link SkytellsClient.config | config.prefetchMaxSlugs}
+ *   (smaller when {@link ClientOptions.runtime} is `"edge"`).
+ * - {@link SkytellsClient.wait} / `run(..., onProgress)` use `setTimeout` between polls; pass {@link WaitOptions.signal}
+ *   (or `RunOptions.signal` with `onProgress`) to abort and stop scheduling further delays.
+ * - {@link Prediction} holds the shared {@link HTTP} client and a snapshot of the API response (expected).
+ * - {@link SkytellsClient.queue} is an unbounded in-memory array — call {@link SkytellsClient.dispatch} or avoid piling
+ *   requests if you need a hard limit (enforce in your app).
+ * - Timeouts, stream cleanup, and polling guards: see **Reliability.md** in the repo docs.
+ *
+ * @module client
+ */
+const { HTTP, HTTP_DEFAULT_REQUEST_TIMEOUT_MS } = require("./http.js");
+const { ENDPOINTS, ORCHESTRATOR_BASE_URL } = require("./endpoints.js");
+const { Orchestrator } = require("./orchestrator.js");
+const { resolvePredictionResourceUrl } = require("./prediction-urls.js");
+const { Chat } = require("./chat.js");
+const { Responses } = require("./responses.js");
+const { Embeddings } = require("./embeddings.js");
+const { Safety } = require("./safety.js");
+const { PredictionStatus } = require("./types/index.js");
+const { Webhook, createWebhookListener } = require("./webhooks.js");
+const { SkytellsError } = require("./types/shared.types.js");
+const DEFAULT_POLL_INTERVAL = 5000;
+/**
+ * Time-to-live for in-memory cache of `GET /model/{slug}` responses used when
+ * a guarded predict/create/run uses `compatibilityCheck: true`. After this interval the next guarded request refetches the model.
+ */
+const PREFETCHED_MODEL_CACHE_TTL_MS = module.exports.PREFETCHED_MODEL_CACHE_TTL_MS = 10 * 60 * 1000;
+/**
+ * Max distinct model slugs kept in the compatibility-check cache. Oldest entries are removed (FIFO) when exceeded.
+ * Prevents unbounded growth if callers use many unique `model` strings over time.
+ */
+const PREFETCHED_MODEL_CACHE_MAX_SLUGS = module.exports.PREFETCHED_MODEL_CACHE_MAX_SLUGS = 64;
+/**
+ * Default request timeout when {@link ClientOptions.runtime} is `"edge"` and `timeout` is omitted
+ * (aligns with common ~25–30s serverless / edge ceilings).
+ */
+const EDGE_DEFAULT_REQUEST_TIMEOUT_MS = module.exports.EDGE_DEFAULT_REQUEST_TIMEOUT_MS = 25000;
+/** Max model slugs in the inference-compat cache when `runtime` is `"edge"` (lower memory). */
+const EDGE_PREFETCH_MAX_SLUGS = module.exports.EDGE_PREFETCH_MAX_SLUGS = 16;
+let edgeRuntimeHintLogged = false;
+let orchestratorKeyFormatWarned = false;
+function logEdgeRuntimeHints(params) {
+    if (edgeRuntimeHintLogged) {
+        return;
+    }
+    edgeRuntimeHintLogged = true;
+    const lines = [
+        '[skytells] ClientOptions.runtime "edge" — edge-oriented defaults and tips:',
+        `  • Request timeout: ${params.requestTimeoutMs}ms${params.userSetTimeout
+            ? ''
+            : ' (edge default when timeout omitted; pass timeout in options to override)'}.`,
+        `  • Inference compat cache: max ${params.prefetchMaxSlugs} slug(s).`,
+        '  • For wait()/run(…, onProgress): pass AbortSignal and maxWait to stay within wall-clock limits.',
+        '  • Webhook HMAC: Web Crypto (crypto.subtle) only.',
+    ];
+    if (params.retryRetries > 2) {
+        lines.push(`  • retry.retries=${params.retryRetries} stacks delay on failures — keep low on edge.`);
+    }
+    if (params.userSetTimeout && params.requestTimeoutMs > 30000) {
+        lines.push(`  • timeout ${params.requestTimeoutMs}ms may exceed typical edge limits (~25–30s); consider lowering.`);
+    }
+    console.warn(lines.join('\n'));
+}
+const DEPRECATED_PREDICTION_METHOD_HINTS = {
+    streamPrediction: 'Use prediction.stream() on the Prediction instance returned from client.run(), or pass prediction response urls when needed.',
+    cancelPrediction: 'Use prediction.cancel() on the Prediction from client.run().',
+    deletePrediction: 'Use prediction.delete() on the Prediction from client.run().',
+};
+function deprecateSkytellsClientPredictionMethod(name) {
+    console.warn(`[skytells] SkytellsClient.${name}() is deprecated and will be removed in a future major version. ${DEPRECATED_PREDICTION_METHOD_HINTS[name]}`);
+}
+/**
+ * Terminal prediction statuses — polling stops when the prediction reaches one of these.
+ */
+const TERMINAL_STATUSES = new Set([
+    PredictionStatus.SUCCEEDED,
+    PredictionStatus.FAILED,
+    PredictionStatus.CANCELLED,
+]);
+/** JSON body for `POST /predict` — normalizes {@link Webhook} to plain `webhook`. */
+function predictionBodyForHttp(payload, overrides) {
+    const { webhook, ...rest } = payload;
+    const webhookJson = webhook === undefined ? undefined : webhook instanceof Webhook ? webhook.toJSON() : webhook;
+    return {
+        ...rest,
+        ...(webhookJson ? { webhook: webhookJson } : {}),
+        ...overrides,
+    };
+}
+/**
+ * Represents a completed prediction with convenience methods for accessing output
+ * and managing the prediction lifecycle (cancel, delete).
+ *
+ * Returned by {@link SkytellsClient.run}. Wraps a {@link PredictionResponse} and
+ * exposes `.output` to get results as a normalized `string[]`, plus `.cancel()`
+ * and `.delete()` for lifecycle management.
+ *
+ * @example
+ * ```ts
+ * const prediction = await client.run("flux-pro", { input: { prompt: "a cat" } });
+ * const [imageUrl] = prediction.output;
+ * console.log(prediction.id, prediction.status);
+ * const streamMeta = await prediction.stream(); // stream endpoint info (uses urls.stream)
+ * // Clean up when done
+ * await prediction.delete();
+ * ```
+ */
+export class Prediction {
+    /** @internal Use {@link SkytellsClient.run} to create a Prediction. */
+    constructor(http, data) {
+        this.http = http;
+        this.data = data;
+    }
+    /**
+     * The full prediction response object from the API.
+     * Contains all fields: status, id, input, output, metrics, metadata, urls, etc.
+     */
+    get response() {
+        return this.data;
+    }
+    /**
+     * The unique prediction identifier.
+     * Use this to fetch, cancel, or delete the prediction later via the client.
+     */
+    get id() {
+        return this.data.id;
+    }
+    /**
+     * The current status of the prediction.
+     * Possible values: `pending`, `starting`, `started`, `processing`, `succeeded`, `failed`, `cancelled`.
+     */
+    get status() {
+        return this.data.status;
+    }
+    /**
+     * The raw prediction output as returned by the API.
+     *
+     * Can be a single `string`, a `string[]`, or `undefined` if incomplete.
+     *
+     * @example
+     * ```ts
+     * // Single output (string)
+     * console.log(prediction.output); // "https://..."
+     *
+     * // Multiple outputs (string[])
+     * console.log(prediction.output[0]); // "https://..."
+     *
+     * // Check type
+     * if (Array.isArray(prediction.output)) {
+     *   for (const item of prediction.output) { console.log(item); }
+     * }
+     * ```
+     */
+    get output() {
+        return this.data.output;
+    }
+    /**
+     * Returns the output, normalized to a single value:
+     *
+     * - `undefined` / `null` → `undefined`
+     * - `"https://..."` → `"https://..."`
+     * - `["https://..."]` → `"https://..."` (single-element array unwrapped)
+     * - `["a", "b"]` → `["a", "b"]` (multi-element array kept as-is)
+     *
+     * @example
+     * ```ts
+     * const result = prediction.outputs();
+     * // string if single output, string[] if multiple, undefined if none
+     * ```
+     */
+    outputs() {
+        if (!this.data.output) {
+            return undefined;
+        }
+        if (typeof this.data.output === 'string') {
+            return this.data.output;
+        }
+        if (this.data.output.length === 1) {
+            return this.data.output[0];
+        }
+        return this.data.output;
+    }
+    /**
+     * Returns the full raw prediction response as a plain JSON object.
+     *
+     * @returns The raw `PredictionResponse` data from the API.
+     *
+     * @example
+     * ```ts
+     * const prediction = await client.run("flux-pro", { input: { prompt: "a cat" } });
+     * const json = prediction.raw();
+     * console.log(json.id, json.status, json.output, json.metrics);
+     * ```
+     */
+    raw() {
+        return this.data;
+    }
+    /**
+     * Fetches streaming metadata for this prediction (`GET` using `urls.stream` when present).
+     * Same behavior as the deprecated {@link SkytellsClient.streamPrediction} with this prediction’s id and urls.
+     *
+     * @returns The prediction response from the stream-info endpoint (includes `urls`, etc.).
+     */
+    async stream() {
+        const url = resolvePredictionResourceUrl('stream', this.data.id, this.data.urls, this.http.getBaseUrl());
+        return this.http.request('GET', url);
+    }
+    /**
+     * Cancels this prediction. Only works if the prediction is still in progress
+     * (status is `pending`, `starting`, `started`, or `processing`).
+     *
+     * @returns The updated prediction response with `status: 'cancelled'`.
+     * @throws {SkytellsError} If the prediction cannot be cancelled or the request fails.
+     *
+     * @example
+     * ```ts
+     * const prediction = await client.run("flux-pro", { input: { prompt: "a cat" } });
+     * await prediction.cancel();
+     * ```
+     */
+    async cancel() {
+        const url = resolvePredictionResourceUrl('cancel', this.data.id, this.data.urls, this.http.getBaseUrl());
+        return this.http.request('POST', url);
+    }
+    /**
+     * Deletes this prediction and its associated output/assets from storage.
+     *
+     * @returns The prediction response confirming deletion.
+     * @throws {SkytellsError} If the prediction cannot be deleted or the request fails.
+     *
+     * @example
+     * ```ts
+     * const prediction = await client.run("flux-pro", { input: { prompt: "a cat" } });
+     * const [url] = prediction.output;
+     * // ... use the output ...
+     * await prediction.delete(); // clean up
+     * ```
+     */
+    async delete() {
+        const url = resolvePredictionResourceUrl('delete', this.data.id, this.data.urls, this.http.getBaseUrl());
+        return this.http.request('DELETE', url);
+    }
+}
+/**
+ * Sub-API for managing predictions. Accessible via `client.predictions`.
+ *
+ * Provides methods to create predictions in the background, fetch them by ID,
+ * and list them with filters.
+ *
+ * @example
+ * ```ts
+ * // Create a background prediction
+ * const prediction = await client.predictions.create({
+ *   model: "flux-pro",
+ *   input: { prompt: "A sunset" },
+ * });
+ *
+ * // Fetch it later
+ * const result = await client.predictions.get(prediction.id);
+ *
+ * // List recent predictions by model
+ * const { data } = await client.predictions.list({ model: "flux-pro", since: "2026-01-01" });
+ * ```
+ */
+export class PredictionsAPI {
+    /** @internal */
+    constructor(http, onBeforePredict) {
+        this.http = http;
+        this.onBeforePredict = onBeforePredict;
+    }
+    /**
+     * Creates a prediction in the background (does not wait for completion).
+     *
+     * The prediction starts processing asynchronously. Use {@link get} to poll,
+     * or {@link SkytellsClient.wait} to block until it finishes.
+     *
+     * @param payload - The prediction request parameters.
+     * @param payload.model - The model slug (e.g. `"flux-pro"`).
+     * @param payload.input - Key-value input parameters for the model.
+     * @param payload.webhook - Optional webhook to receive events.
+     * @param payload.stream - Enable streaming (default: `false`).
+     * @param sdk - SDK-only options (not sent in JSON). When `sdk.compatibilityCheck === true`, runs inference compatibility guard before POST.
+     * @returns The initial prediction response with `status: 'pending'` or `'starting'`.
+     * @throws {SkytellsError} On API errors (invalid input, model not found, insufficient credits).
+     *
+     * @example
+     * ```ts
+     * const prediction = await client.predictions.create({
+     *   model: "flux-pro",
+     *   input: { prompt: "An astronaut riding a unicorn" },
+     * });
+     * console.log(prediction.id, prediction.status); // "pending"
+     *
+     * // Wait for it to finish
+     * const result = await client.wait(prediction);
+     * console.log(result.output);
+     * ```
+     */
+    async create(payload, sdk) {
+        if (this.onBeforePredict) {
+            await this.onBeforePredict(payload, sdk);
+        }
+        return this.http.request('POST', ENDPOINTS.PREDICT, predictionBodyForHttp(payload, { await: false }));
+    }
+    /**
+     * Fetches a prediction by its ID.
+     *
+     * @param id - The prediction ID.
+     * @param urls - Optional `urls` from a prior `PredictionResponse`. When `urls.get` is set, that absolute URL is used (API protocol); otherwise `GET {baseUrl}/predictions/{id}`.
+     * @returns The prediction response with current status, output, metrics, etc.
+     * @throws {SkytellsError} If the prediction is not found or the request fails.
+     *
+     * @example
+     * ```ts
+     * const prediction = await client.predictions.get("pred_abc123");
+     * if (prediction.status === "succeeded") {
+     *   console.log(prediction.output);
+     * }
+     * ```
+     */
+    async get(id, urls) {
+        const url = resolvePredictionResourceUrl('get', id, urls, this.http.getBaseUrl());
+        return this.http.request('GET', url);
+    }
+    /**
+     * Lists predictions with optional filters and pagination.
+     *
+     * @param options - Filter and pagination options.
+     * @param options.page - Page number (default: 1).
+     * @param options.since - Only include predictions created on or after this date (`YYYY-MM-DD`).
+     * @param options.until - Only include predictions created on or before this date (`YYYY-MM-DD`).
+     * @param options.model - Filter by model slug.
+     * @returns A paginated response with `data` (predictions array) and `pagination` metadata.
+     * @throws {SkytellsError} On authentication failure or invalid parameters.
+     *
+     * @example
+     * ```ts
+     * // List all predictions
+     * const { data, pagination } = await client.predictions.list();
+     *
+     * // Filter by model and date range
+     * const filtered = await client.predictions.list({
+     *   model: "flux-pro",
+     *   since: "2026-01-01",
+     *   until: "2026-03-15",
+     *   page: 2,
+     * });
+     * ```
+     */
+    async list(options) {
+        const params = new URLSearchParams();
+        if (options === null || options === void 0 ? void 0 : options.page) {
+            params.set('page', String(options.page));
+        }
+        if (options === null || options === void 0 ? void 0 : options.since) {
+            params.set('from', options.since);
+        }
+        if (options === null || options === void 0 ? void 0 : options.until) {
+            params.set('to', options.until);
+        }
+        if (options === null || options === void 0 ? void 0 : options.model) {
+            params.set('model', options.model);
+        }
+        const query = params.toString();
+        const path = query ? `${ENDPOINTS.PREDICTIONS}?${query}` : ENDPOINTS.PREDICTIONS;
+        return this.http.request('GET', path);
+    }
+}
+/**
+ * Sub-API for browsing and fetching models. Accessible via `client.models`.
+ *
+ * Provides `.list()` to get all models and `.get()` to fetch a single model by slug.
+ *
+ * @example
+ * ```ts
+ * const allModels = await client.models.list();
+ * const model = await client.models.get("flux-pro", { fields: ["input_schema"] });
+ * ```
+ */
+export class ModelsAPI {
+    /** @internal */
+    constructor(http) {
+        this.http = http;
+    }
+    /**
+     * Lists all available models on the Skytells platform.
+     *
+     * @param options - Optional configuration.
+     * @param options.fields - Additional fields to include in the response
+     *   (e.g. `["input_schema", "output_schema"]`). By default, schemas are not included.
+     * @returns An array of model objects with name, type, pricing, metadata, etc.
+     * @throws {SkytellsError} On authentication failure or server error.
+     *
+     * @example
+     * ```ts
+     * const allModels = await client.models.list();
+     * for (const m of allModels) {
+     *   console.log(m.name, m.type);
+     * }
+     *
+     * // Include input schemas
+     * const withSchemas = await client.models.list({ fields: ["input_schema"] });
+     * ```
+     */
+    async list(options) {
+        var _a;
+        let path = ENDPOINTS.MODELS;
+        if ((_a = options === null || options === void 0 ? void 0 : options.fields) === null || _a === void 0 ? void 0 : _a.length) {
+            path += `?fields=${options.fields.join(',')}`;
+        }
+        return this.http.request('GET', path);
+    }
+    /**
+     * Fetches a single model by its slug.
+     *
+     * @param slug - The model slug identifier (e.g. `"flux-pro"`, `"truefusion"`, `"beatfusion"`).
+     * @param options - Optional configuration.
+     * @param options.fields - Additional fields to include (e.g. `["input_schema", "output_schema"]`).
+     * @returns The model object with name, type, pricing, metadata, and optionally schemas.
+     * @throws {SkytellsError} If the model is not found (`MODEL_NOT_FOUND`) or the request fails.
+     *
+     * @example
+     * ```ts
+     * const model = await client.models.get("flux-pro");
+     * console.log(model.name, model.pricing);
+     *
+     * // With schemas
+     * const detailed = await client.models.get("flux-pro", {
+     *   fields: ["input_schema", "output_schema"],
+     * });
+     * console.log(detailed.input_schema);
+     * ```
+     */
+    async get(slug, options) {
+        var _a;
+        let path = ENDPOINTS.MODEL_BY_SLUG(slug);
+        if ((_a = options === null || options === void 0 ? void 0 : options.fields) === null || _a === void 0 ? void 0 : _a.length) {
+            path += `?fields=${options.fields.join(',')}`;
+        }
+        return this.http.request('GET', path);
+    }
+}
+/**
+ * The main Skytells API client. Provides methods to run predictions, list models,
+ * and manage prediction lifecycle.
+ *
+ * Create an instance using {@link Skytells} or the constructor directly.
+ *
+ * @example
+ * ```ts
+ * const Skytells = require("skytells");
+ *
+ * const client = Skytells("sk-your-api-key", {
+ *   timeout: 30000,
+ *   retry: { retries: 2 },
+ * });
+ *
+ * // Run a model and get output
+ * const prediction = await client.run("flux-pro", {
+ *   input: { prompt: "An astronaut riding a unicorn" },
+ * });
+ * const [imageUrl] = prediction.output;
+ *
+ * // List and fetch models
+ * const allModels = await client.models.list();
+ * const model = await client.models.get("flux-pro");
+ *
+ * // Background prediction
+ * const bg = await client.predictions.create({
+ *   model: "flux-pro",
+ *   input: { prompt: "A sunset" },
+ * });
+ * const result = await client.wait(bg);
+ *
+ * // Queue multiple predictions
+ * client.queue({ model: "flux-pro", input: { prompt: "Cat" } });
+ * client.queue({ model: "flux-pro", input: { prompt: "Dog" } });
+ * const results = await client.dispatch();
+ * ```
+ */
 export class SkytellsClient {
     /**
-     * Creates a new Skytells client
-     * @param apiKey Your Skytells API key
-     * @param options Configuration options
+     * Sub-API for managing predictions.
+     * Provides `.create()`, `.get()`, and `.list()` methods.
+     */
+    get predictions() {
+        if (!this._predictions) {
+            this._predictions = new PredictionsAPI(this.http, (p, sdk) => this._maybeCompatibilityGuard(p.model, sdk === null || sdk === void 0 ? void 0 : sdk.compatibilityCheck));
+        }
+        return this._predictions;
+    }
+    /**
+     * Alias of {@link predictions} (singular). Same instance — use whichever reads best.
+     *
+     * @example
+     * ```ts
+     * await client.prediction.create({ model: 'flux-pro', input: { prompt: '…' } });
+     * // same as client.predictions.create(...)
+     * ```
+     */
+    get prediction() {
+        return this.predictions;
+    }
+    /**
+     * Sub-API for browsing and fetching models.
+     * Provides `.list()` and `.get()` methods.
+     */
+    get models() {
+        if (!this._models) {
+            this._models = new ModelsAPI(this.http);
+        }
+        return this._models;
+    }
+    /**
+     * Chat completions. Same as OpenAI's client.chat.completions.
+     *
+     * @example
+     * ```ts
+     * const completion = await client.chat.completions.create({
+     *   model: 'deepbrain-router',
+     *   messages: [{ role: 'user', content: 'Hello' }],
+     * });
+     * ```
+     */
+    get chat() {
+        if (!this._chat) {
+            this._chat = new Chat(this.http);
+        }
+        return this._chat;
+    }
+    /**
+     * Responses API (`POST /v1/responses`). Same as `client.chat.responses`.
+     *
+     * Follows the OpenAI Responses API schema. Supports both non-streaming and streaming modes.
+     *
+     * @example Non-streaming:
+     * ```ts
+     * const response = await client.responses.create({
+     *   model: 'gpt-5.3-codex',
+     *   input: [{ role: 'user', content: 'Explain recursion simply.' }],
+     *   instructions: 'You are a helpful tutor.',
+     * });
+     * console.log(response.output[0].content[0].text);
+     * ```
+     *
+     * @example Streaming:
+     * ```ts
+     * const stream = await client.responses.create({
+     *   model: 'gpt-5.3-codex',
+     *   input: [{ role: 'user', content: 'Explain recursion simply.' }],
+     *   stream: true,
+     * });
+     * for await (const event of stream) {
+     *   console.log(event.type, event);
+     * }
+     * ```
+     */
+    get responses() {
+        if (!this._responses) {
+            this._responses = new Responses(this.http);
+        }
+        return this._responses;
+    }
+    /**
+     * Embeddings. Same as OpenAI's client.embeddings.
+     *
+     * @example
+     * ```ts
+     * const embedding = await client.embeddings.create({
+     *   model: 'text-embedding-3-small',
+     *   input: 'Hello world',
+     * });
+     * ```
+     */
+    get embeddings() {
+        if (!this._embeddings) {
+            this._embeddings = new Embeddings(this.http);
+        }
+        return this._embeddings;
+    }
+    /**
+     * Safety checks. Proactive (checkText, checkImage) and response parsing (wasFiltered, evaluate).
+     * evaluate() accepts text, image URLs, choices, completions, prediction results, or arrays of any.
+     *
+     * @example
+     * ```ts
+     * const result = await client.safety.checkText('user input');
+     * if (client.safety.wasFiltered(completion)) { ... }
+     * const evalResult = await client.safety.evaluate('user text', SafetyTemplates.STRICT);
+     * ```
+     */
+    get safety() {
+        if (!this._safety) {
+            this._safety = new Safety(this.http);
+        }
+        return this._safety;
+    }
+    /**
+     * [Skytells Orchestrator](https://learn.skytells.ai/docs/products/orchestrator/api-reference) — workflows,
+     * executions, webhook triggers, integrations, etc.
+     *
+     * Requires **`ClientOptions.orchestratorApiKey`** (`wfb_…`). Uses an internal HTTP stack aimed at the Orchestrator
+     * host ({@link ORCHESTRATOR_BASE_URL}): Bearer only, no `x-api-key`. Same `timeout`, `headers`, `retry`, and
+     * `fetch` as the main Skytells client — pass both keys on one {@link SkytellsClient} when you need both products.
+     *
+     * @throws {SkytellsError} `SDK_ERROR` if `orchestratorApiKey` was not set.
+     */
+    get orchestrator() {
+        if (!this._orchestratorApiKey) {
+            throw new SkytellsError('client.orchestrator requires ClientOptions.orchestratorApiKey (Orchestrator wfb_… key). It is not your Skytells sk-… platform key.', 'SDK_ERROR', 'https://learn.skytells.ai/docs/products/orchestrator/api-keys', 0);
+        }
+        if (!this._orchestrator) {
+            const b = this._orchestratorHttpBundle;
+            this._orchestrator = new Orchestrator(new HTTP(this._orchestratorApiKey, this._orchestratorBaseUrl, b.timeout, b.headers, b.retry, b.fetch, 'orchestrator'));
+        }
+        return this._orchestrator;
+    }
+    /**
+     * Target runtime from {@link ClientOptions.runtime} (`"default"` when omitted).
+     */
+    get runtime() {
+        return this._runtime;
+    }
+    /**
+     * Read-only resolved settings (timeouts, cache cap) for debugging and tests.
+     */
+    get config() {
+        return {
+            runtime: this._runtime,
+            requestTimeoutMs: this._requestTimeoutMs,
+            prefetchMaxSlugs: this._prefetchCap,
+        };
+    }
+    /**
+     * Creates a new Skytells API client.
+     *
+     * @param apiKey - Your Skytells API key (starts with `sk-`). Required for authenticated endpoints.
+     * @param options - Client configuration options.
+     * @param options.baseUrl - Override the default API base URL.
+     * @param options.timeout - Request timeout in milliseconds (default: 60000, or 25000 when `runtime: "edge"` and omitted).
+     * @param options.headers - Custom headers to include in every request.
+     * @param options.retry - Retry configuration: `retries`, `retryDelay`, `retryOn` status codes.
+     * @param options.fetch - Custom `fetch` implementation (useful for testing, proxying, or
+     *   disabling Next.js fetch caching).
+     * @param options.runtime - `"edge"` for Workers / Vercel Edge / etc.: shorter default timeout, smaller compat cache, one-time hints.
+     *
+     * @example
+     * ```ts
+     * const client = new SkytellsClient("sk-your-api-key");
+     *
+     * // With options:
+     * const client = new SkytellsClient("sk-your-api-key", {
+     *   timeout: 30000,
+     *   retry: { retries: 3, retryDelay: 1000 },
+     *   headers: { "X-Custom-Header": "value" },
+     * });
+     *
+     * // Next.js App Router — disable fetch caching:
+     * const client = new SkytellsClient("sk-your-api-key", {
+     *   fetch: (url, opts) => fetch(url, { ...opts, cache: "no-store" }),
+     * });
+     *
+     * // Edge / serverless:
+     * const edge = new SkytellsClient("sk-…", { runtime: "edge" });
+     *
+     * // Skytells + Orchestrator (optional wfb_… key — same client):
+     * const both = new SkytellsClient("sk-…", { orchestratorApiKey: "wfb_…" });
+     * ```
      */
     constructor(apiKey, options = {}) {
-        this.http = new HTTP(apiKey, options.baseUrl, options.timeout);
+        var _a, _b, _c, _d, _e, _f, _g;
+        this._queue = [];
+        /**
+         * In-memory `GET /model/{slug}` results for the per-prediction guard. Lazy-allocated; bounded by
+         * {@link SkytellsClient.config | config.prefetchMaxSlugs} and TTL {@link PREFETCHED_MODEL_CACHE_TTL_MS}.
+         */
+        this._prefetchedModelCache = null;
+        const runtime = (_a = options.runtime) !== null && _a !== void 0 ? _a : 'default';
+        const isEdge = runtime === 'edge';
+        const userSetTimeout = options.timeout !== undefined;
+        const requestTimeoutMs = userSetTimeout
+            ? options.timeout
+            : isEdge
+                ? EDGE_DEFAULT_REQUEST_TIMEOUT_MS
+                : HTTP_DEFAULT_REQUEST_TIMEOUT_MS;
+        const prefetchCap = isEdge ? EDGE_PREFETCH_MAX_SLUGS : PREFETCHED_MODEL_CACHE_MAX_SLUGS;
+        this._runtime = runtime;
+        this._requestTimeoutMs = requestTimeoutMs;
+        this._prefetchCap = prefetchCap;
+        if (isEdge) {
+            logEdgeRuntimeHints({
+                requestTimeoutMs,
+                retryRetries: (_c = (_b = options.retry) === null || _b === void 0 ? void 0 : _b.retries) !== null && _c !== void 0 ? _c : 0,
+                userSetTimeout,
+                prefetchMaxSlugs: prefetchCap,
+            });
+        }
+        this.http = new HTTP(apiKey, options.baseUrl, requestTimeoutMs, options.headers, options.retry, options.fetch);
+        this._orchestratorApiKey = ((_d = options.orchestratorApiKey) === null || _d === void 0 ? void 0 : _d.trim()) || undefined;
+        if (this._orchestratorApiKey && !this._orchestratorApiKey.startsWith('wfb_')) {
+            if (!orchestratorKeyFormatWarned) {
+                orchestratorKeyFormatWarned = true;
+                console.warn('[skytells] orchestratorApiKey normally starts with "wfb_" (Orchestrator webhook key, separate from sk-…). See https://learn.skytells.ai/docs/products/orchestrator/api-keys');
+            }
+        }
+        this._orchestratorBaseUrl = ((_e = options.orchestratorBaseUrl) === null || _e === void 0 ? void 0 : _e.trim()) || ORCHESTRATOR_BASE_URL;
+        this._orchestratorHttpBundle = {
+            timeout: requestTimeoutMs,
+            headers: (_f = options.headers) !== null && _f !== void 0 ? _f : {},
+            retry: (_g = options.retry) !== null && _g !== void 0 ? _g : {},
+            fetch: options.fetch,
+        };
+    }
+    /**
+     * Build a {@link WebhookListener} for your HTTP server. Verifies `X-Skytells-Signature` per
+     * [Skytells webhooks](https://docs.skytells.ai/webhooks/).
+     *
+     * - **`mode: 'general'`** (default): HMAC key is your API key — omitted `apiKey` uses this client’s key.
+     * - **`mode: 'enterprise'`**: pass `secret` from the dashboard.
+     *
+     * @example
+     * ```ts
+     * const hooks = client.webhookListener({ mode: 'general' });
+     * hooks.on(WebhookEvent.COMPLETED, async (p) => console.log(p.id));
+     * // Next.js: export async function POST(req: Request) { return hooks.handleRequest(req); }
+     * ```
+     */
+    webhookListener(options = {}) {
+        var _a;
+        const apiKey = (_a = options.apiKey) !== null && _a !== void 0 ? _a : this.http.getApiKey();
+        return createWebhookListener({ ...options, apiKey });
+    }
+    /**
+     * Alias of {@link SkytellsClient.webhookListener}. Chain handlers: `client.listen().on(WebhookEvent.COMPLETED, fn)`.
+     */
+    listen(options) {
+        return this.webhookListener(options);
+    }
+    /**
+     * Clears cached model metadata used when guarded predict/create/run passes `compatibilityCheck: true`.
+     *
+     * @param modelSlug - If set, only evicts that slug; otherwise clears the entire cache.
+     *
+     * @example
+     * ```ts
+     * // After a model was updated server-side, force the next predict to refetch metadata:
+     * client.purgePrefetchedModelCache("flux-pro");
+     *
+     * // Clear all cached slugs
+     * client.purgePrefetchedModelCache();
+     * ```
+     */
+    purgePrefetchedModelCache(modelSlug) {
+        const cache = this._prefetchedModelCache;
+        if (!cache) {
+            return;
+        }
+        if (modelSlug !== undefined) {
+            cache.delete(modelSlug);
+            if (cache.size === 0) {
+                this._prefetchedModelCache = null;
+            }
+        }
+        else {
+            this._prefetchedModelCache = null;
+        }
+    }
+    /**
+     * Returns model metadata for the inference compatibility check, using {@link _prefetchedModelCache} when fresh.
+     */
+    async _getModelForCompatibilityCheck(slug) {
+        const now = Date.now();
+        if (!this._prefetchedModelCache) {
+            this._prefetchedModelCache = new Map();
+        }
+        const cache = this._prefetchedModelCache;
+        const cached = cache.get(slug);
+        if (cached && cached.expiresAt > now) {
+            // Move to end of Map iteration order so FIFO eviction keeps recently used slugs.
+            cache.delete(slug);
+            cache.set(slug, cached);
+            return cached.model;
+        }
+        if (cached) {
+            cache.delete(slug);
+        }
+        const model = await this.models.get(slug);
+        cache.set(slug, {
+            model,
+            expiresAt: now + PREFETCHED_MODEL_CACHE_TTL_MS,
+        });
+        while (cache.size > this._prefetchCap) {
+            const oldest = cache.keys().next().value;
+            if (oldest === undefined) {
+                break;
+            }
+            cache.delete(oldest);
+        }
+        return model;
+    }
+    /**
+     * @internal When `compatibilityCheck === true`, throws SDK_ERROR if the model is chat-only here.
+     */
+    async _maybeCompatibilityGuard(modelSlug, compatibilityCheck) {
+        var _a;
+        if (compatibilityCheck !== true) {
+            return;
+        }
+        try {
+            const modelData = await this._getModelForCompatibilityCheck(modelSlug);
+            if ((_a = modelData.metadata) === null || _a === void 0 ? void 0 : _a.openai_compatible) {
+                throw new SkytellsError(`This model supports the OpenAI Chat Completions API. Did you mean client.chat.completions.create()? Use it for chat-style inference.`, 'SDK_ERROR', `Model "${modelData.namespace}" is OpenAI-compatible. Use client.chat.completions.create() instead of predict() or run().`, 0);
+            }
+        }
+        catch (e) {
+            if (e instanceof SkytellsError && e.errorId === 'SDK_ERROR') {
+                throw e;
+            }
+            // If model fetch fails (e.g. MODEL_NOT_FOUND), let the predict request proceed
+        }
+    }
+    /**
+     * Sends a prediction request to the Skytells API.
+     *
+     * This is the low-level prediction method. For most use cases, prefer {@link run}
+     * which waits for completion and returns a {@link Prediction} object with convenience methods.
+     *
+     * @param payload - The prediction request parameters.
+     * @param payload.model - The model slug to run (e.g. `"flux-pro"`, `"truefusion"`).
+     * @param payload.input - Key-value input parameters for the model (e.g. `{ prompt: "..." }`).
+     * @param payload.await - If `true`, the request blocks until the prediction completes (default: `false`).
+     * @param payload.stream - If `true`, enables streaming for the prediction (default: `false`).
+     * @param payload.webhook - Optional webhook configuration to receive prediction events.
+     * @param sdk - SDK-only options (not sent in JSON). When `sdk.compatibilityCheck === true`, runs inference compatibility guard before POST.
+     * @returns The prediction response object.
+     * @throws {SkytellsError} On API errors (invalid input, model not found, insufficient credits, etc.).
+     *
+     * @example
+     * ```ts
+     * // Fire-and-forget (returns immediately with status: "pending")
+     * const response = await client.predict({
+     *   model: "flux-pro",
+     *   input: { prompt: "A sunset over mountains" },
+     * });
+     * console.log(response.id); // use predictions.get(id) to poll
+     *
+     * // Wait for completion
+     * const result = await client.predict({
+     *   model: "flux-pro",
+     *   input: { prompt: "A sunset over mountains" },
+     *   await: true,
+     * });
+     * console.log(result.output); // ["https://..."]
+     * ```
+     */
+    async predict(payload, sdk) {
+        await this._maybeCompatibilityGuard(payload.model, sdk === null || sdk === void 0 ? void 0 : sdk.compatibilityCheck);
+        return this.http.request('POST', ENDPOINTS.PREDICT, predictionBodyForHttp(payload));
+    }
+    /**
+     * Runs a model, waits for completion, and returns a {@link Prediction} object.
+     *
+     * This is the recommended way to generate content. It automatically sets `await: true`,
+     * checks for failures, and returns a `Prediction` with `.output`, `.cancel()`, and `.delete()`.
+     *
+     * If an `onProgress` callback is provided, the prediction is created in the background
+     * and polled every 5 seconds. The callback is invoked on each poll with the latest
+     * prediction state, allowing you to track progress.
+     *
+     * @param model - The model slug to run (e.g. `"flux-pro"`, `"truefusion"`, `"beatfusion"`).
+     * @param options - Run options.
+     * @param options.input - Key-value input parameters for the model.
+     * @param options.stream - If `true`, enables streaming (default: `false`).
+     * @param options.webhook - Optional webhook configuration.
+     * @param onProgress - Optional callback invoked on each poll with the current prediction state.
+     * @param sdk - SDK-only options. When `sdk.compatibilityCheck === true`, runs inference guard before predict/create. Use `undefined` for `onProgress` when you only need `sdk` (e.g. `run(m, o, undefined, { compatibilityCheck: true })`).
+     * @remarks With `onProgress`, uses `predictions.create` + {@link wait}. Then {@link RunOptions.interval},
+     *   {@link RunOptions.maxWait}, and {@link RunOptions.signal} apply (same as {@link WaitOptions}).
+     * @returns A {@link Prediction} object with output and lifecycle methods.
+     * @throws {SkytellsError} On API errors or if the prediction fails (`PREDICTION_FAILED`).
+     *
+     * @example
+     * ```ts
+     * // Simple run (blocks until complete)
+     * const prediction = await client.run("flux-pro", {
+     *   input: { prompt: "An astronaut riding a rainbow unicorn" },
+     * });
+     * const [imageUrl] = prediction.output;
+     *
+     * // With progress tracking
+     * const prediction = await client.run("flux-pro", {
+     *   input: { prompt: "An astronaut riding a unicorn" },
+     * }, (p) => {
+     *   // Note: metrics.progress may not be available during early processing stages
+     *   console.log(`Status: ${p.status}, Progress: ${p.metrics?.progress ?? 'n/a'}`);
+     * });
+     * ```
+     */
+    async run(model, options, onProgress, sdk) {
+        const { input, webhook, stream, interval, maxWait, signal } = options;
+        const predictionBody = { input, webhook, stream };
+        let data;
+        if (onProgress) {
+            data = await this.predictions.create({ model, ...predictionBody }, sdk);
+            data = await this.wait(data, { interval, maxWait, signal }, onProgress);
+        }
+        else {
+            data = await this.predict({ model, ...predictionBody, await: true }, sdk);
+        }
+        if (data.status === PredictionStatus.FAILED) {
+            throw new SkytellsError(data.response || 'Prediction failed', 'PREDICTION_FAILED', `Prediction ${data.id} failed`);
+        }
+        return new Prediction(this.http, data);
+    }
+    /**
+     * Polls a prediction until it reaches a terminal status (`succeeded`, `failed`, or `cancelled`).
+     *
+     * @param prediction - The prediction response to wait on (must have an `id`).
+     * @param options - Polling options.
+     * @param options.interval - Polling interval in milliseconds (default: 5000).
+     * @param options.maxWait - Maximum wait time in milliseconds. Throws if exceeded.
+     * @param options.signal - If aborted, stops polling and throws `ABORTED` (no further poll timers scheduled).
+     * @param onProgress - Optional callback invoked on each poll with the latest prediction state.
+     * @returns The final prediction response.
+     * @throws {SkytellsError} If `maxWait` is exceeded (`WAIT_TIMEOUT`) or `signal` is aborted (`ABORTED`).
+     *
+     * @example
+     * ```ts
+     * const bg = await client.predictions.create({
+     *   model: "flux-pro",
+     *   input: { prompt: "A cat" },
+     * });
+     *
+     * const result = await client.wait(bg);
+     * console.log(result.output);
+     *
+     * // With progress and timeout
+     * const result = await client.wait(bg, { interval: 2000, maxWait: 120000 }, (p) => {
+     *   console.log(p.status, p.metrics?.progress);
+     * });
+     * ```
+     */
+    async wait(prediction, options, onProgress) {
+        var _a;
+        const id = prediction === null || prediction === void 0 ? void 0 : prediction.id;
+        if (id == null || id === '') {
+            throw new SkytellsError('Cannot wait on a prediction without an id', 'SDK_ERROR', 'PredictionResponse.id is required for polling', 0);
+        }
+        const rawInterval = (_a = options === null || options === void 0 ? void 0 : options.interval) !== null && _a !== void 0 ? _a : DEFAULT_POLL_INTERVAL;
+        const interval = typeof rawInterval === 'number' && Number.isFinite(rawInterval) && rawInterval >= 0
+            ? rawInterval
+            : DEFAULT_POLL_INTERVAL;
+        const rawMax = options === null || options === void 0 ? void 0 : options.maxWait;
+        const maxWaitMs = typeof rawMax === 'number' && Number.isFinite(rawMax) && rawMax >= 0 ? rawMax : undefined;
+        const signal = options === null || options === void 0 ? void 0 : options.signal;
+        const startTime = Date.now();
+        const deadline = typeof maxWaitMs === 'number' ? startTime + maxWaitMs : null;
+        let current = prediction;
+        if (TERMINAL_STATUSES.has(current.status)) {
+            return current;
+        }
+        let firstPoll = true;
+        while (!TERMINAL_STATUSES.has(current.status)) {
+            if (signal === null || signal === void 0 ? void 0 : signal.aborted) {
+                throw SkytellsClient.abortWaitError();
+            }
+            const now = Date.now();
+            if (deadline !== null && now >= deadline) {
+                throw new SkytellsError(`Prediction ${current.id} did not complete within ${maxWaitMs}ms`, 'WAIT_TIMEOUT', `Timed out after ${maxWaitMs}ms. Last status: ${current.status}`, 408);
+            }
+            if (!firstPoll) {
+                const remaining = deadline !== null ? deadline - Date.now() : interval;
+                if (deadline !== null && remaining <= 0) {
+                    throw new SkytellsError(`Prediction ${current.id} did not complete within ${maxWaitMs}ms`, 'WAIT_TIMEOUT', `Timed out after ${maxWaitMs}ms. Last status: ${current.status}`, 408);
+                }
+                const sleepMs = Math.min(interval, Math.max(0, remaining));
+                await this.delay(sleepMs, signal);
+                if (signal === null || signal === void 0 ? void 0 : signal.aborted) {
+                    throw SkytellsClient.abortWaitError();
+                }
+                if (deadline !== null && Date.now() >= deadline) {
+                    throw new SkytellsError(`Prediction ${current.id} did not complete within ${maxWaitMs}ms`, 'WAIT_TIMEOUT', `Timed out after ${maxWaitMs}ms. Last status: ${current.status}`, 408);
+                }
+            }
+            firstPoll = false;
+            current = await this.predictions.get(current.id, current.urls);
+            if (onProgress) {
+                onProgress(current);
+            }
+        }
+        return current;
+    }
+    /**
+     * Adds a prediction request to the local queue.
+     *
+     * Queued items are not sent to the API until {@link dispatch} is called.
+     * Useful for batching multiple predictions together.
+     *
+     * @param payload - The prediction request to queue.
+     * @param sdk - Optional SDK options for this item (e.g. `{ compatibilityCheck: true }`), forwarded to {@link PredictionsAPI.create} on {@link dispatch}.
+     *
+     * @example
+     * ```ts
+     * client.queue({ model: "flux-pro", input: { prompt: "Cat" } });
+     * client.queue({ model: "flux-pro", input: { prompt: "Dog" } });
+     * client.queue({ model: "flux-pro", input: { prompt: "Bird" } });
+     *
+     * const results = await client.dispatch();
+     * // results is an array of PredictionResponse for each queued item
+     * ```
+     */
+    queue(payload, sdk) {
+        this._queue.push({ request: payload, sdk });
+    }
+    /**
+     * Dispatches all queued predictions, sending them to the API concurrently.
+     *
+     * Clears the queue after dispatching. Each prediction is created in the background
+     * (non-blocking). Use {@link wait} on individual results if you need to wait for completion.
+     *
+     * @returns An array of prediction responses, one for each queued item.
+     * @throws {SkytellsError} Uses **`Promise.all`**: if **any** `predictions.create` rejects, the whole
+     *   `dispatch()` rejects with that error (fail-fast). Partial successes are not returned.
+     *
+     * @example
+     * ```ts
+     * client.queue({ model: "flux-pro", input: { prompt: "Cat" } });
+     * client.queue({ model: "flux-pro", input: { prompt: "Dog" } });
+     *
+     * const results = await client.dispatch();
+     * for (const pred of results) {
+     *   console.log(pred.id, pred.status);
+     * }
+     *
+     * // Wait for all to complete
+     * const completed = await Promise.all(results.map(p => client.wait(p)));
+     * ```
+     */
+    async dispatch() {
+        const items = this._queue.splice(0);
+        const results = await Promise.all(items.map((item) => this.predictions.create(item.request, item.sdk)));
+        return results;
     }
     /**
-     * Send a prediction request to the Skytells API
-     * @param payload The prediction request parameters
-     * @returns A promise that resolves to the prediction response
+     * Retrieves the streaming endpoint for a prediction.
+     *
+     * @deprecated Use {@link Prediction.stream} on the {@link Prediction} from {@link SkytellsClient.run} instead,
+     * or pass `urls` from a stored `PredictionResponse` when calling without a `Prediction` instance.
+     *
+     * @param id - The prediction ID to stream.
+     * @param urls - Optional `urls` from a prior `PredictionResponse`. When `urls.stream` is set, that URL is used.
+     * @returns The prediction response with streaming URL info.
+     * @throws {SkytellsError} If the prediction is not found or streaming is not available.
+     *
+     * @example
+     * ```ts
+     * const stream = await client.streamPrediction("pred_abc123");
+     * console.log(stream.urls?.stream);
+     * ```
      */
-    async predict(payload) {
-        return this.http.request('POST', ENDPOINTS.PREDICT, payload);
+    async streamPrediction(id, urls) {
+        deprecateSkytellsClientPredictionMethod('streamPrediction');
+        const url = resolvePredictionResourceUrl('stream', id, urls, this.http.getBaseUrl());
+        return this.http.request('GET', url);
+    }
+    /**
+     * Cancels a running prediction by its ID.
+     *
+     * @deprecated Use {@link Prediction.cancel} on the {@link Prediction} from {@link SkytellsClient.run} instead.
+     * Only predictions with status `pending`, `starting`, `started`, or `processing` can be cancelled.
+     *
+     * @param id - The prediction ID to cancel.
+     * @param urls - Optional `urls` from a prior `PredictionResponse`. When `urls.cancel` is set, that URL is used.
+     * @returns The updated prediction response with `status: 'cancelled'`.
+     * @throws {SkytellsError} If the prediction cannot be cancelled or is not found.
+     *
+     * @example
+     * ```ts
+     * await client.cancelPrediction("pred_abc123");
+     * ```
+     */
+    async cancelPrediction(id, urls) {
+        deprecateSkytellsClientPredictionMethod('cancelPrediction');
+        const url = resolvePredictionResourceUrl('cancel', id, urls, this.http.getBaseUrl());
+        return this.http.request('POST', url);
+    }
+    /**
+     * Deletes a prediction and its associated output/assets.
+     *
+     * @deprecated Use {@link Prediction.delete} on the {@link Prediction} from {@link SkytellsClient.run} instead.
+     *
+     * @param id - The prediction ID to delete.
+     * @param urls - Optional `urls` from a prior `PredictionResponse`. When `urls.delete` is set, that URL is used.
+     * @returns The prediction response confirming deletion.
+     * @throws {SkytellsError} If the prediction is not found or the request fails.
+     *
+     * @example
+     * ```ts
+     * await client.deletePrediction("pred_abc123");
+     * ```
+     */
+    async deletePrediction(id, urls) {
+        deprecateSkytellsClientPredictionMethod('deletePrediction');
+        const url = resolvePredictionResourceUrl('delete', id, urls, this.http.getBaseUrl());
+        return this.http.request('DELETE', url);
+    }
+    /** Stops {@link wait} polling when `signal` aborts; clears the pending timer so nothing keeps firing. */
+    static abortWaitError() {
+        return new SkytellsError('Wait aborted', 'ABORTED', 'Polling stopped because the AbortSignal was aborted.', 0);
     }
     /**
-     * Get a prediction by ID
-     * @param id The prediction ID
-     * @returns A promise that resolves to the prediction response
+     * @param signal - When present and aborted, the promise rejects with `ABORTED` and the timer is cleared.
      */
-    async getPrediction(id) {
-        return this.http.request('GET', ENDPOINTS.PREDICTION_BY_ID(id));
+    delay(ms, signal) {
+        const d = Math.max(0, ms);
+        if (!signal) {
+            return new Promise((resolve) => setTimeout(resolve, d));
+        }
+        if (signal.aborted) {
+            return Promise.reject(SkytellsClient.abortWaitError());
+        }
+        return new Promise((resolve, reject) => {
+            let settled = false;
+            const finish = (fn) => {
+                if (settled) {
+                    return;
+                }
+                settled = true;
+                clearTimeout(tid);
+                signal.removeEventListener('abort', onAbort);
+                fn();
+            };
+            const onAbort = () => {
+                finish(() => reject(SkytellsClient.abortWaitError()));
+            };
+            signal.addEventListener('abort', onAbort);
+            const tid = setTimeout(() => {
+                finish(() => resolve());
+            }, d);
+        });
     }
+    // ── Deprecated aliases ──────────────────────────────────────────────
     /**
-     * List all available models
-     * @returns A promise that resolves to an array of models
+     * @deprecated Use {@link predictions}.list() instead. This method will be removed in a future version.
      */
-    async listModels() {
-        return this.http.request('GET', ENDPOINTS.MODELS);
+    async listPredictions(page) {
+        return this.predictions.list({ page });
     }
     /**
-     * Stream a prediction by ID
-     * @param id The prediction ID
-     * @returns A promise that resolves to the prediction response
+     * @deprecated Use {@link models}.list() instead. This method will be removed in a future version.
      */
-    async streamPrediction(id) {
-        return this.http.request('GET', ENDPOINTS.STREAM_PREDICTION_BY_ID(id));
+    async listModels(options) {
+        return this.models.list(options);
     }
     /**
-     * Cancel a prediction by ID
-     * @param id The prediction ID
-     * @returns A promise that resolves to the prediction response
+     * @deprecated Use {@link predictions}.get() instead. This method will be removed in a future version.
      */
-    async cancelPrediction(id) {
-        return this.http.request('POST', ENDPOINTS.CANCEL_PREDICTION_BY_ID(id));
+    async getPrediction(id, urls) {
+        return this.predictions.get(id, urls);
     }
     /**
-     * Delete a prediction by ID
-     * @param id The prediction ID
-     * @returns A promise that resolves to the prediction response
+     * @deprecated Use {@link models}.get() instead. This method will be removed in a future version.
      */
-    async deletePrediction(id) {
-        return this.http.request('DELETE', ENDPOINTS.DELETE_PREDICTION_BY_ID(id));
+    async getModel(slug, options) {
+        return this.models.get(slug, options);
     }
 }