@promptbook/remote-client/esm/typings/src/execution/LlmExecutionTools.d.ts

Promptbook: Turn your company's scattered knowledge into AI ready books

import type { Promisable } from 'type-fest';
import type { ChatParticipant } from '../book-components/Chat/types/ChatParticipant';
import type { Prompt } from '../types/Prompt';
import type { string_markdown, string_markdown_text, string_title } from '../types/typeAliases';
import type { AvailableModel } from './AvailableModel';
import type { ChatPromptResult, CompletionPromptResult, EmbeddingPromptResult } from './PromptResult';
/**
 * Container for all the tools needed to execute prompts to large language models like GPT-4
 * On its interface it exposes common methods for prompt execution.
 * Inside (in constructor) it calls OpenAI, Azure, GPU, proxy, cache, logging,...
 *
 * Note: [🦖] There are several different things in Promptbook:
 * - `Agent` - which represents an AI Agent with its source, memories, actions, etc. Agent is a higher-level abstraction which is internally using:
 * - `LlmExecutionTools` - which wraps one or more LLM models and provides an interface to execute them
 * - `AgentLlmExecutionTools` - which is a specific implementation of `LlmExecutionTools` that wraps another LlmExecutionTools and applies agent-specific system prompts and requirements
 * - `OpenAiAssistantExecutionTools` - which is a specific implementation of `LlmExecutionTools` for OpenAI models with assistant capabilities, recommended for usage in `Agent` or `AgentLlmExecutionTools`
 * - `RemoteAgent` - which is an `Agent` that connects to a Promptbook Agents Server
 *
 * @see https://github.com/webgptorg/promptbook#llm-execution-tools
 */
export type LlmExecutionTools = {
    /**
     * Title of the model provider
     *
     * @example "OpenAI"
     */
    readonly title: string_title & string_markdown_text;
    /**
     * Description of the provider
     *
     * @example "Use all models from OpenAI"
     */
    readonly description?: string_markdown;
    /**
     * Profile representing the model as a virtual persona for chat interfaces
     * This includes visual identity like avatar, colors, and display information
     *
     * @example { name: 'OPENAI', fullname: 'OpenAI GPT', color: '#10a37f', avatarSrc: '/openai-logo.png' }
     */
    readonly profile?: ChatParticipant;
    /**
     * Check comfiguration
     *
     * @returns nothing if configuration is correct
     * @throws {Error} if configuration is incorrect
     */
    checkConfiguration(): Promisable<void>;
    /**
     * List all available models that can be used
     */
    listModels(): Promisable<ReadonlyArray<AvailableModel>>;
    /**
     * Calls a chat model
     */
    callChatModel?(prompt: Prompt): Promise<ChatPromptResult>;
    /**
     * Calls a chat model with streaming
     */
    callChatModelStream?(prompt: Prompt, onProgress: (chunk: ChatPromptResult) => void): Promise<ChatPromptResult>;
    /**
     * Calls a completion model
     */
    callCompletionModel?(prompt: Prompt): Promise<CompletionPromptResult>;
    /**
     * Calls an embedding model
     */
    callEmbeddingModel?(prompt: Prompt): Promise<EmbeddingPromptResult>;
};
/**
 * TODO: [🕛] Extend this from sth class -  like `AgentBasicInformation` / `ModelBasicInformation``
 * TODO: [🍚] Implement destroyable pattern to free resources
 * TODO: [🏳] Add `callTranslationModel`
 * TODO: [🧠] Emulation of one type of model with another one - emuate chat with completion; emulate translation with chat
 * TODO: [🍓][♐] Some heuristic to pick the best model in listed models
 * TODO: [🧠] Should or should not there be a word "GPT" in both callCompletionModel and callChatModel
 * TODO: [🧠][🪐] Should be common things like types, utils in folder containing A,B,C,.. or else outside this listing folder?
 */