@prismshadow/mmsp 0.5.0

package/dist/autoClient.d.ts

@@ -0,0 +1,96 @@
+import { LLMClient } from "./baseClient";
+import { UniConfig, UniEvent, UniMessage } from "./types";
+export declare const OFFICIAL_CLIENT_TYPES: readonly ["openai-official", "anthropic-official", "gemini-official", "zai-official", "moonshot-official", "deepseek-official", "minimax-official"];
+export declare const COMPATIBLE_CLIENT_TYPES: readonly ["openai-responses", "openai-chat", "openai-chat-vllm-adapter", "openai-embedding", "ant-messages", "gemini-generate-content"];
+export declare const MODEL_FAMILIES: [string, string][];
+/**
+ * The official client a model id routes to on its own, or null when its family is unknown.
+ *
+ * @param model - The model id, in any casing
+ * @returns One of OFFICIAL_CLIENT_TYPES, or null
+ */
+export declare function clientTypeForModel(model: string): string | null;
+/**
+ * The one client to call: it creates the client a client type names and forwards to it.
+ *
+ * A client type is one of OFFICIAL_CLIENT_TYPES or COMPATIBLE_CLIENT_TYPES, given as
+ * `clientType` or as the CLIENT_TYPE environment variable. Without one, the family the model
+ * id begins with names its official client: `gpt-` routes to `openai-official`, `claude-` to
+ * `anthropic-official`, and so on. A model id of no known family throws, and asks for a
+ * client type.
+ *
+ * This client is stateful - it knows the model name at initialization and maintains
+ * conversation history for that specific model.
+ */
+export declare class AutoLLMClient extends LLMClient {
+    private _client;
+    private _clientType;
+    private _named;
+    /**
+     * Initialize AutoLLMClient with a model and, unless the model id names it, a client type.
+     *
+     * @param options - Configuration object with model, apiKey, baseUrl, clientType and defaultHeaders
+     */
+    constructor(options: {
+        model: string;
+        apiKey?: string;
+        baseUrl?: string | null;
+        clientType?: string | null;
+        defaultHeaders?: Record<string, string>;
+    });
+    /**
+     * Delegate to underlying client's transformUniConfigToModelConfig.
+     */
+    transformUniConfigToModelConfig(config: UniConfig): any;
+    /**
+     * Delegate to underlying client's transformUniMessageToModelInput.
+     */
+    transformUniMessageToModelInput(messages: UniMessage[], signal?: AbortSignal): any;
+    /**
+     * Delegate to underlying client's transformModelOutputToUniEvent.
+     */
+    transformModelOutputToUniEvent(modelOutput: any): UniEvent;
+    /**
+     * Not implemented - use streamingResponse instead.
+     */
+    _streamingResponseInternal(_options: any): AsyncGenerator<UniEvent>;
+    /**
+     * Route to underlying client's streamingResponse.
+     */
+    streamingResponse(options: {
+        messages: UniMessage[];
+        config: UniConfig;
+        signal?: AbortSignal;
+    }): AsyncGenerator<UniEvent>;
+    /**
+     * Route to underlying client's streamingResponseStateful.
+     */
+    streamingResponseStateful(options: {
+        message: UniMessage;
+        config: UniConfig;
+        signal?: AbortSignal;
+    }): AsyncGenerator<UniEvent>;
+    /**
+     * Clear history in the underlying client.
+     */
+    clearHistory(): void;
+    /**
+     * Get history from the underlying client.
+     */
+    getHistory(): UniMessage[];
+    /**
+     * Set history in the underlying client.
+     */
+    setHistory(history: UniMessage[]): void;
+    /**
+     * List the model ids the endpoint serves that this client can be used for.
+     *
+     * A client named by its type speaks for whatever the endpoint serves, so its listing is
+     * returned whole. A client deduced from a model id serves the ids that deduce to it as well,
+     * so a gateway fronting many vendors is filtered down to that client's own models.
+     *
+     * @returns The model ids, in the order the endpoint returned them.
+     */
+    listModels(): Promise<string[]>;
+}
+//# sourceMappingURL=autoClient.d.ts.map

package/dist/autoClient.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"autoClient.d.ts","sourceRoot":"","sources":["../src/autoClient.ts"],"names":[],"mappings":"AAcA,OAAO,EAAE,SAAS,EAAE,MAAM,cAAc,CAAC;AAczC,OAAO,EAAE,SAAS,EAAE,QAAQ,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AAW1D,eAAO,MAAM,qBAAqB,qJAQxB,CAAC;AAGX,eAAO,MAAM,uBAAuB,yIAO1B,CAAC;AAGX,eAAO,MAAM,cAAc,EAAE,CAAC,MAAM,EAAE,MAAM,CAAC,EAS5C,CAAC;AAEF;;;;;GAKG;AACH,wBAAgB,kBAAkB,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAQ/D;AA6DD;;;;;;;;;;;GAWG;AACH,qBAAa,aAAc,SAAQ,SAAS;IAC1C,OAAO,CAAC,OAAO,CAAY;IAC3B,OAAO,CAAC,WAAW,CAAS;IAE5B,OAAO,CAAC,MAAM,CAAU;IAExB;;;;OAIG;gBACS,OAAO,EAAE;QACnB,KAAK,EAAE,MAAM,CAAC;QACd,MAAM,CAAC,EAAE,MAAM,CAAC;QAChB,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;QACxB,UAAU,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;QAC3B,cAAc,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;KACzC;IA4BD;;OAEG;IAEH,+BAA+B,CAAC,MAAM,EAAE,SAAS,GAAG,GAAG;IAIvD;;OAEG;IAEH,+BAA+B,CAC7B,QAAQ,EAAE,UAAU,EAAE,EACtB,MAAM,CAAC,EAAE,WAAW,GACnB,GAAG;IAKN;;OAEG;IAEH,8BAA8B,CAAC,WAAW,EAAE,GAAG,GAAG,QAAQ;IAI1D;;OAEG;IAEI,0BAA0B,CAAC,QAAQ,EAAE,GAAG,GAAG,cAAc,CAAC,QAAQ,CAAC;IAI1E;;OAEG;IACI,iBAAiB,CAAC,OAAO,EAAE;QAChC,QAAQ,EAAE,UAAU,EAAE,CAAC;QACvB,MAAM,EAAE,SAAS,CAAC;QAClB,MAAM,CAAC,EAAE,WAAW,CAAC;KACtB,GAAG,cAAc,CAAC,QAAQ,CAAC;IAU5B;;OAEG;IACI,yBAAyB,CAAC,OAAO,EAAE;QACxC,OAAO,EAAE,UAAU,CAAC;QACpB,MAAM,EAAE,SAAS,CAAC;QAClB,MAAM,CAAC,EAAE,WAAW,CAAC;KACtB,GAAG,cAAc,CAAC,QAAQ,CAAC;IAU5B;;OAEG;IACH,YAAY,IAAI,IAAI;IAIpB;;OAEG;IACH,UAAU,IAAI,UAAU,EAAE;IAI1B;;OAEG;IACH,UAAU,CAAC,OAAO,EAAE,UAAU,EAAE,GAAG,IAAI;IAIvC;;;;;;;;OAQG;IACG,UAAU,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;CAStC"}

package/dist/autoClient.js

@@ -0,0 +1,258 @@
+"use strict";
+// Copyright 2025 Prism Shadow. and/or its affiliates
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.AutoLLMClient = exports.MODEL_FAMILIES = exports.COMPATIBLE_CLIENT_TYPES = exports.OFFICIAL_CLIENT_TYPES = void 0;
+exports.clientTypeForModel = clientTypeForModel;
+const baseClient_1 = require("./baseClient");
+const gemini_official_1 = require("./gemini_official");
+const gemini_generate_content_1 = require("./gemini_generate_content");
+const anthropic_official_1 = require("./anthropic_official");
+const openai_official_1 = require("./openai_official");
+const zai_official_1 = require("./zai_official");
+const moonshot_official_1 = require("./moonshot_official");
+const openai_chat_1 = require("./openai_chat");
+const openai_responses_1 = require("./openai_responses");
+const ant_messages_1 = require("./ant_messages");
+const openai_embedding_1 = require("./openai_embedding");
+const deepseek_official_1 = require("./deepseek_official");
+const minimax_official_1 = require("./minimax_official");
+const openai_chat_vllm_adapter_1 = require("./openai_chat_vllm_adapter");
+// An official client speaks its vendor's own API, knows the vendor's models, and reads the
+// vendor's key from the environment.
+exports.OFFICIAL_CLIENT_TYPES = [
+    "openai-official",
+    "anthropic-official",
+    "gemini-official",
+    "zai-official",
+    "moonshot-official",
+    "deepseek-official",
+    "minimax-official",
+];
+// A compatible client speaks one wire protocol for whatever endpoint serves it.
+exports.COMPATIBLE_CLIENT_TYPES = [
+    "openai-responses",
+    "openai-chat",
+    "openai-chat-vllm-adapter",
+    "openai-embedding",
+    "ant-messages",
+    "gemini-generate-content",
+];
+// Without a client type, the family a model id begins with names its official client.
+exports.MODEL_FAMILIES = [
+    ["gpt-", "openai-official"],
+    ["text-embedding-", "openai-official"],
+    ["claude-", "anthropic-official"],
+    ["gemini-", "gemini-official"],
+    ["glm-", "zai-official"],
+    ["kimi-", "moonshot-official"],
+    ["deepseek-", "deepseek-official"],
+    ["minimax-", "minimax-official"],
+];
+/**
+ * The official client a model id routes to on its own, or null when its family is unknown.
+ *
+ * @param model - The model id, in any casing
+ * @returns One of OFFICIAL_CLIENT_TYPES, or null
+ */
+function clientTypeForModel(model) {
+    const lowered = model.toLowerCase();
+    for (const [prefix, clientType] of exports.MODEL_FAMILIES) {
+        if (lowered.startsWith(prefix)) {
+            return clientType;
+        }
+    }
+    return null;
+}
+function clientTypes() {
+    return (`official clients: ${exports.OFFICIAL_CLIENT_TYPES.join(", ")}; ` +
+        `compatible clients: ${exports.COMPATIBLE_CLIENT_TYPES.join(", ")}`);
+}
+/**
+ * The client class a client type names, or null when no client is named so.
+ *
+ * @param clientType - A lowercased client type
+ * @param model - The model id, which tells an official client's embedding models apart
+ * @param apiKey - The key, which tells a Vertex AI service account apart
+ */
+function clientClass(clientType, model, apiKey) {
+    switch (clientType) {
+        case "openai-official":
+            // OpenAI serves embedding models through its Embeddings API
+            return model.toLowerCase().startsWith("text-embedding-")
+                ? openai_embedding_1.OpenaiEmbeddingClient
+                : openai_official_1.OpenAIOfficialClient;
+        case "anthropic-official":
+            return anthropic_official_1.AnthropicOfficialClient;
+        case "gemini-official":
+            // Vertex AI serves none of the Gemini models through the Interactions API, so a
+            // service-account JSON key, which only Vertex AI takes, speaks generateContent
+            return (apiKey || process.env.GEMINI_API_KEY || "").startsWith("{")
+                ? gemini_generate_content_1.GeminiGenerateContentClient
+                : gemini_official_1.GeminiOfficialClient;
+        case "zai-official":
+            return zai_official_1.ZAIOfficialClient;
+        case "moonshot-official":
+            return moonshot_official_1.MoonshotOfficialClient;
+        case "deepseek-official":
+            return deepseek_official_1.DeepSeekOfficialClient;
+        case "minimax-official":
+            return minimax_official_1.MiniMaxOfficialClient;
+        case "openai-responses":
+            return openai_responses_1.OpenaiResponsesClient;
+        case "openai-chat":
+        case "openai":
+            return openai_chat_1.OpenaiChatClient;
+        case "openai-chat-vllm-adapter":
+            return openai_chat_vllm_adapter_1.OpenaiChatVllmAdapterClient;
+        case "openai-embedding":
+            return openai_embedding_1.OpenaiEmbeddingClient;
+        case "ant-messages":
+            return ant_messages_1.AntMessagesClient;
+        case "gemini-generate-content":
+            return gemini_generate_content_1.GeminiGenerateContentClient;
+        default:
+            return null;
+    }
+}
+/**
+ * The one client to call: it creates the client a client type names and forwards to it.
+ *
+ * A client type is one of OFFICIAL_CLIENT_TYPES or COMPATIBLE_CLIENT_TYPES, given as
+ * `clientType` or as the CLIENT_TYPE environment variable. Without one, the family the model
+ * id begins with names its official client: `gpt-` routes to `openai-official`, `claude-` to
+ * `anthropic-official`, and so on. A model id of no known family throws, and asks for a
+ * client type.
+ *
+ * This client is stateful - it knows the model name at initialization and maintains
+ * conversation history for that specific model.
+ */
+class AutoLLMClient extends baseClient_1.LLMClient {
+    /**
+     * Initialize AutoLLMClient with a model and, unless the model id names it, a client type.
+     *
+     * @param options - Configuration object with model, apiKey, baseUrl, clientType and defaultHeaders
+     */
+    constructor(options) {
+        super();
+        const named = options.clientType || process.env.CLIENT_TYPE;
+        const clientType = named
+            ? named.toLowerCase()
+            : clientTypeForModel(options.model);
+        if (clientType === null) {
+            throw new Error(`No client for model ${JSON.stringify(options.model)}: its family is not known. ` +
+                `Pass clientType, one of the ${clientTypes()}.`);
+        }
+        const ClientClass = clientClass(clientType, options.model, options.apiKey);
+        if (ClientClass === null) {
+            throw new Error(`Unknown client type ${JSON.stringify(named)}. Pass one of the ${clientTypes()}.`);
+        }
+        this._clientType = clientType;
+        this._named = Boolean(named);
+        this._client = new ClientClass({
+            model: options.model,
+            apiKey: options.apiKey,
+            baseUrl: options.baseUrl,
+            defaultHeaders: options.defaultHeaders,
+        });
+    }
+    /**
+     * Delegate to underlying client's transformUniConfigToModelConfig.
+     */
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    transformUniConfigToModelConfig(config) {
+        return this._client.transformUniConfigToModelConfig(config);
+    }
+    /**
+     * Delegate to underlying client's transformUniMessageToModelInput.
+     */
+    /* eslint-disable @typescript-eslint/no-explicit-any */
+    transformUniMessageToModelInput(messages, signal) {
+        return this._client.transformUniMessageToModelInput(messages, signal);
+    }
+    /* eslint-enable @typescript-eslint/no-explicit-any */
+    /**
+     * Delegate to underlying client's transformModelOutputToUniEvent.
+     */
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    transformModelOutputToUniEvent(modelOutput) {
+        return this._client.transformModelOutputToUniEvent(modelOutput);
+    }
+    /**
+     * Not implemented - use streamingResponse instead.
+     */
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any, require-yield
+    async *_streamingResponseInternal(_options) {
+        throw new Error("Please use streamingResponse instead.");
+    }
+    /**
+     * Route to underlying client's streamingResponse.
+     */
+    async *streamingResponse(options) {
+        for await (const event of this._client.streamingResponse({
+            messages: options.messages,
+            config: options.config,
+            signal: options.signal,
+        })) {
+            yield event;
+        }
+    }
+    /**
+     * Route to underlying client's streamingResponseStateful.
+     */
+    async *streamingResponseStateful(options) {
+        for await (const event of this._client.streamingResponseStateful({
+            message: options.message,
+            config: options.config,
+            signal: options.signal,
+        })) {
+            yield event;
+        }
+    }
+    /**
+     * Clear history in the underlying client.
+     */
+    clearHistory() {
+        this._client.clearHistory();
+    }
+    /**
+     * Get history from the underlying client.
+     */
+    getHistory() {
+        return this._client.getHistory();
+    }
+    /**
+     * Set history in the underlying client.
+     */
+    setHistory(history) {
+        this._client.setHistory(history);
+    }
+    /**
+     * List the model ids the endpoint serves that this client can be used for.
+     *
+     * A client named by its type speaks for whatever the endpoint serves, so its listing is
+     * returned whole. A client deduced from a model id serves the ids that deduce to it as well,
+     * so a gateway fronting many vendors is filtered down to that client's own models.
+     *
+     * @returns The model ids, in the order the endpoint returned them.
+     */
+    async listModels() {
+        const modelIds = await this._client.listModels();
+        if (this._named) {
+            return modelIds;
+        }
+        return modelIds.filter((modelId) => clientTypeForModel(modelId) === this._clientType);
+    }
+}
+exports.AutoLLMClient = AutoLLMClient;

package/dist/baseClient.d.ts

@@ -0,0 +1,111 @@
+import { UniConfig, UniEvent, UniMessage } from "./types";
+/**
+ * Abstract base class for LLM clients.
+ *
+ * All model-specific clients must inherit from this class and implement
+ * the required abstract methods for complete SDK abstraction.
+ */
+export declare abstract class LLMClient {
+    protected _model: string;
+    private _history;
+    constructor();
+    /**
+     * Transform universal configuration to model-specific configuration.
+     *
+     * @param config - Universal configuration object
+     * @returns Model-specific configuration object
+     */
+    abstract transformUniConfigToModelConfig(config: UniConfig): any;
+    /**
+     * Transform universal message format to model-specific input format.
+     *
+     * @param messages - List of universal message objects
+     * @returns Model-specific input format (e.g., Gemini's Content list, OpenAI's messages array)
+     */
+    abstract transformUniMessageToModelInput(messages: UniMessage[], signal?: AbortSignal): any;
+    /**
+     * Transform one event of the provider's stream into a universal event, which the base class
+     * turns into the public stream.
+     *
+     * content_items holds the deltas the wire event carries, in wire order, and never a done item:
+     * the base class closes an item when the next one begins or the stream ends. The deltas of one
+     * item are contiguous and carry the same `fidelity.item_id`, the provider's id for the item
+     * where it has one, which never reaches the public stream. Its event_type is "stop" on the wire
+     * events that report usage_metadata and/or finish_reason, in pieces the base class merges field
+     * by field, and "delta" otherwise; a "delta" event carries neither.
+     *
+     * @param modelOutput - Model-specific output object (streaming chunk)
+     * @returns Universal event object, an empty delta event when the wire event carries nothing
+     *   universal
+     */
+    abstract transformModelOutputToUniEvent(modelOutput: any): UniEvent;
+    /**
+     * Concatenate a stream of universal events into a single universal message.
+     *
+     * @param events - List of universal events from streaming response
+     * @returns Complete universal message object: every done item in stream order, with the
+     *   usage, finish reason and timestamp of the stop event
+     */
+    concatUniEventsToUniMessage(events: UniEvent[]): UniMessage;
+    /**
+     * Internal method to handle streaming response.
+     *
+     * Each model client implements it to send the request and yield one universal event per event
+     * of the provider's stream; streamingResponse turns them into the public stream.
+     *
+     * @param options - Object containing messages and config
+     * @yields Universal events of the streaming response
+     */
+    abstract _streamingResponseInternal(options: {
+        messages: UniMessage[];
+        config: UniConfig;
+        signal?: AbortSignal;
+    }): AsyncGenerator<UniEvent>;
+    /**
+     * List the model ids the configured endpoint serves.
+     *
+     * @returns The model ids, in the order the endpoint returned them.
+     */
+    abstract listModels(): Promise<string[]>;
+    /**
+     * Generate content in streaming mode (stateless).
+     *
+     * @param options - Object containing messages and config
+     * @yields Delta events, each carrying one delta or done item, then exactly one stop event
+     *   carrying the usage and the finish reason
+     */
+    streamingResponse(options: {
+        messages: UniMessage[];
+        config: UniConfig;
+        signal?: AbortSignal;
+    }): AsyncGenerator<UniEvent>;
+    /**
+     * Generate content in streaming mode (stateful).
+     *
+     * @param message - Latest universal message object to add to conversation
+     * @param config - Universal configuration object
+     * @yields Universal events from the streaming response
+     */
+    streamingResponseStateful(options: {
+        message: UniMessage;
+        config: UniConfig;
+        signal?: AbortSignal;
+    }): AsyncGenerator<UniEvent>;
+    /**
+     * Clear the message history.
+     */
+    clearHistory(): void;
+    /**
+     * Get the current message history.
+     *
+     * @returns Copy of the current message history
+     */
+    getHistory(): UniMessage[];
+    /**
+     * Replace the message history with a copy of the provided history.
+     *
+     * @param history - List of universal message objects to set as the new history
+     */
+    setHistory(history: UniMessage[]): void;
+}
+//# sourceMappingURL=baseClient.d.ts.map

package/dist/baseClient.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"baseClient.d.ts","sourceRoot":"","sources":["../src/baseClient.ts"],"names":[],"mappings":"AAiBA,OAAO,EAKL,SAAS,EACT,QAAQ,EACR,UAAU,EAEX,MAAM,SAAS,CAAC;AAuHjB;;;;;GAKG;AACH,8BAAsB,SAAS;IAC7B,SAAS,CAAC,MAAM,EAAE,MAAM,CAAC;IACzB,OAAO,CAAC,QAAQ,CAAe;;IAO/B;;;;;OAKG;IAEH,QAAQ,CAAC,+BAA+B,CAAC,MAAM,EAAE,SAAS,GAAG,GAAG;IAEhE;;;;;OAKG;IACH,QAAQ,CAAC,+BAA+B,CACtC,QAAQ,EAAE,UAAU,EAAE,EACtB,MAAM,CAAC,EAAE,WAAW,GACnB,GAAG;IAEN;;;;;;;;;;;;;;OAcG;IAEH,QAAQ,CAAC,8BAA8B,CAAC,WAAW,EAAE,GAAG,GAAG,QAAQ;IAEnE;;;;;;OAMG;IACH,2BAA2B,CAAC,MAAM,EAAE,QAAQ,EAAE,GAAG,UAAU;IAwB3D;;;;;;;;OAQG;IACH,QAAQ,CAAC,0BAA0B,CAAC,OAAO,EAAE;QAC3C,QAAQ,EAAE,UAAU,EAAE,CAAC;QACvB,MAAM,EAAE,SAAS,CAAC;QAClB,MAAM,CAAC,EAAE,WAAW,CAAC;KACtB,GAAG,cAAc,CAAC,QAAQ,CAAC;IAE5B;;;;OAIG;IACH,QAAQ,CAAC,UAAU,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;IAExC;;;;;;OAMG;IACI,iBAAiB,CAAC,OAAO,EAAE;QAChC,QAAQ,EAAE,UAAU,EAAE,CAAC;QACvB,MAAM,EAAE,SAAS,CAAC;QAClB,MAAM,CAAC,EAAE,WAAW,CAAC;KACtB,GAAG,cAAc,CAAC,QAAQ,CAAC;IA4C5B;;;;;;OAMG;IACI,yBAAyB,CAAC,OAAO,EAAE;QACxC,OAAO,EAAE,UAAU,CAAC;QACpB,MAAM,EAAE,SAAS,CAAC;QAClB,MAAM,CAAC,EAAE,WAAW,CAAC;KACtB,GAAG,cAAc,CAAC,QAAQ,CAAC;IAoB5B;;OAEG;IACH,YAAY,IAAI,IAAI;IAIpB;;;;OAIG;IACH,UAAU,IAAI,UAAU,EAAE;IAI1B;;;;OAIG;IACH,UAAU,CAAC,OAAO,EAAE,UAAU,EAAE,GAAG,IAAI;CAGxC"}