@promptbook/core

v0.112.0-24

Published

6 hours ago

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

0High
0Medium
0Low

hejny

✨ Promptbook: AI Agents

Create persistent AI agents that turn your company's scattered knowledge into action — powered by the Agents Server

🌟 New Features

Gemini 3 Support

📦 Package `@promptbook/core`

Promptbooks are divided into several packages, all are published from single monorepo.
This package @promptbook/core is one part of the promptbook ecosystem.

To install this package, run:

# Install entire promptbook ecosystem
npm i ptbk

# Install just this package to save space
npm install @promptbook/core

The core package contains the fundamental logic and infrastructure for Promptbook. It provides the essential building blocks for creating, parsing, validating, and executing promptbooks, along with comprehensive error handling, LLM provider integrations, and execution utilities.

🎯 Purpose and Motivation

The core package serves as the foundation of the Promptbook ecosystem. It abstracts away the complexity of working with different LLM providers, provides a unified interface for prompt execution, and handles all the intricate details of pipeline management, parameter validation, and result processing.

🔧 High-Level Functionality

This package orchestrates the entire promptbook execution lifecycle:

Pipeline Management: Parse, validate, and compile promptbook definitions
Execution Engine: Create and manage pipeline executors with comprehensive error handling
LLM Integration: Unified interface for multiple LLM providers (OpenAI, Anthropic, Google, etc.)
Parameter Processing: Template parameter substitution and validation
Knowledge Management: Handle knowledge sources and scraping
Storage Abstraction: Flexible storage backends for caching and persistence
Format Support: Parse and validate various data formats (JSON, CSV, XML)

✨ Key Features

🚀 Universal Pipeline Executor - Execute promptbooks with any supported LLM provider
🔄 Multi-Provider Support - Seamlessly switch between OpenAI, Anthropic, Google, and other providers
📊 Comprehensive Validation - Validate promptbooks, parameters, and execution results
🎯 Expectation Checking - Built-in validation for output format, length, and content expectations
🧠 Knowledge Integration - Scrape and process knowledge from various sources
💾 Flexible Storage - Memory, filesystem, and custom storage backends
🔧 Error Handling - Detailed error types for debugging and monitoring
📈 Usage Tracking - Monitor token usage, costs, and performance metrics
🎨 Format Parsers - Support for JSON, CSV, XML, and text formats
🔀 Pipeline Migration - Upgrade and migrate pipeline definitions

📦 Exported Entities

Version Information

BOOK_LANGUAGE_VERSION - Current book language version
PROMPTBOOK_ENGINE_VERSION - Current engine version

Agent and Book Management

createAgentModelRequirements - Create model requirements for agents
parseAgentSource - Parse agent source code
isValidBook - Validate book format
validateBook - Comprehensive book validation
DEFAULT_BOOK - Default book template

Commitment System

createEmptyAgentModelRequirements - Create empty model requirements
createBasicAgentModelRequirements - Create basic model requirements
NotYetImplementedCommitmentDefinition - Placeholder for future commitments
getCommitmentDefinition - Get specific commitment definition
getAllCommitmentDefinitions - Get all available commitment definitions
getAllCommitmentTypes - Get all commitment types
isCommitmentSupported - Check if commitment is supported

Collection Management

pipelineCollectionToJson - Convert collection to JSON
createPipelineCollectionFromJson - Create collection from JSON data
createPipelineCollectionFromPromise - Create collection from async source
createPipelineCollectionFromUrl - Create collection from URL
createPipelineSubcollection - Create filtered subcollection

Configuration Constants

NAME - Project name
ADMIN_EMAIL - Administrator email
ADMIN_GITHUB_NAME - GitHub username
CLAIM - Project claim/tagline
DEFAULT_BOOK_TITLE - Default book title
DEFAULT_TASK_TITLE - Default task title
DEFAULT_PROMPT_TASK_TITLE - Default prompt task title
DEFAULT_BOOK_OUTPUT_PARAMETER_NAME - Default output parameter name
DEFAULT_MAX_FILE_SIZE - Maximum file size limit
BIG_DATASET_TRESHOLD - Threshold for large datasets
FAILED_VALUE_PLACEHOLDER - Placeholder for failed values
PENDING_VALUE_PLACEHOLDER - Placeholder for pending values
MAX_FILENAME_LENGTH - Maximum filename length
DEFAULT_INTERMEDIATE_FILES_STRATEGY - Strategy for intermediate files
DEFAULT_MAX_PARALLEL_COUNT - Maximum parallel executions
DEFAULT_MAX_EXECUTION_ATTEMPTS - Maximum execution attempts
DEFAULT_MAX_KNOWLEDGE_SOURCES_SCRAPING_DEPTH - Knowledge scraping depth limit
DEFAULT_MAX_KNOWLEDGE_SOURCES_SCRAPING_TOTAL - Knowledge scraping total limit
DEFAULT_BOOKS_DIRNAME - Default books directory name
DEFAULT_DOWNLOAD_CACHE_DIRNAME - Default download cache directory
DEFAULT_EXECUTION_CACHE_DIRNAME - Default execution cache directory
DEFAULT_SCRAPE_CACHE_DIRNAME - Default scrape cache directory
CLI_APP_ID - CLI application identifier
PLAYGROUND_APP_ID - Playground application identifier
DEFAULT_PIPELINE_COLLECTION_BASE_FILENAME - Default collection filename
DEFAULT_REMOTE_SERVER_URL - Default remote server URL
DEFAULT_CSV_SETTINGS - Default CSV parsing settings
DEFAULT_IS_VERBOSE - Default verbosity setting
SET_IS_VERBOSE - Verbosity setter
DEFAULT_IS_AUTO_INSTALLED - Default auto-install setting
DEFAULT_TASK_SIMULATED_DURATION_MS - Default task simulation duration
DEFAULT_GET_PIPELINE_COLLECTION_FUNCTION_NAME - Default collection function name
DEFAULT_MAX_REQUESTS_PER_MINUTE - Rate limiting configuration
API_REQUEST_TIMEOUT - API request timeout
PROMPTBOOK_LOGO_URL - Official logo URL

Model and Provider Constants

MODEL_TRUST_LEVELS - Trust levels for different models
MODEL_ORDERS - Ordering preferences for models
ORDER_OF_PIPELINE_JSON - JSON property ordering
RESERVED_PARAMETER_NAMES - Reserved parameter names

Pipeline Processing

compilePipeline - Compile pipeline from source
parsePipeline - Parse pipeline definition
pipelineJsonToString - Convert pipeline JSON to string
prettifyPipelineString - Format pipeline string
extractParameterNamesFromTask - Extract parameter names
validatePipeline - Validate pipeline structure

Dialog and Interface Tools

CallbackInterfaceTools - Callback-based interface tools
CallbackInterfaceToolsOptions - Options for callback tools (type)

Error Handling

BoilerplateError - Base error class
PROMPTBOOK_ERRORS - All error types registry
AbstractFormatError - Abstract format validation error
AuthenticationError - Authentication failure error
CollectionError - Collection-related error
EnvironmentMismatchError - Environment compatibility error
ExpectError - Expectation validation error
KnowledgeScrapeError - Knowledge scraping error
LimitReachedError - Resource limit error
MissingToolsError - Missing tools error
NotFoundError - Resource not found error
NotYetImplementedError - Feature not implemented error
ParseError - Parsing error
PipelineExecutionError - Pipeline execution error
PipelineLogicError - Pipeline logic error
PipelineUrlError - Pipeline URL error
PromptbookFetchError - Fetch operation error
UnexpectedError - Unexpected error
WrappedError - Wrapped error container

Execution Engine

createPipelineExecutor - Create pipeline executor
computeCosineSimilarity - Compute cosine similarity for embeddings
embeddingVectorToString - Convert embedding vector to string
executionReportJsonToString - Convert execution report to string
ExecutionReportStringOptions - Report formatting options (type)
ExecutionReportStringOptionsDefaults - Default report options

Usage and Metrics

addUsage - Add usage metrics
isPassingExpectations - Check if expectations are met
ZERO_VALUE - Zero usage value constant
UNCERTAIN_ZERO_VALUE - Uncertain zero value constant
ZERO_USAGE - Zero usage object
UNCERTAIN_USAGE - Uncertain usage object
usageToHuman - Convert usage to human-readable format
usageToWorktime - Convert usage to work time estimate

Format Parsers

CsvFormatError - CSV format error
CsvFormatParser - CSV format parser
MANDATORY_CSV_SETTINGS - Required CSV settings
TextFormatParser - Text format parser

Form Factor Definitions

BoilerplateFormfactorDefinition - Boilerplate form factor
ChatbotFormfactorDefinition - Chatbot form factor
CompletionFormfactorDefinition - Completion form factor
GeneratorFormfactorDefinition - Generator form factor
GenericFormfactorDefinition - Generic form factor
ImageGeneratorFormfactorDefinition - Image generator form factor
FORMFACTOR_DEFINITIONS - All form factor definitions
MatcherFormfactorDefinition - Matcher form factor
SheetsFormfactorDefinition - Sheets form factor
TranslatorFormfactorDefinition - Translator form factor

LLM Provider Integration

filterModels - Filter available models
$llmToolsMetadataRegister - LLM tools metadata registry
$llmToolsRegister - LLM tools registry
createLlmToolsFromConfiguration - Create tools from config
cacheLlmTools - Cache LLM tools
countUsage - Count total usage
limitTotalUsage - Limit total usage
joinLlmExecutionTools - Join multiple LLM tools
MultipleLlmExecutionTools - Multiple LLM tools container

Provider Registrations

_AnthropicClaudeMetadataRegistration - Anthropic Claude registration
_AzureOpenAiMetadataRegistration - Azure OpenAI registration
_DeepseekMetadataRegistration - Deepseek registration
_GoogleMetadataRegistration - Google registration
_OllamaMetadataRegistration - Ollama registration
_OpenAiMetadataRegistration - OpenAI registration
_OpenAiAssistantMetadataRegistration - OpenAI Assistant registration
_OpenAiCompatibleMetadataRegistration - OpenAI Compatible registration

Pipeline Management

migratePipeline - Migrate pipeline to newer version
preparePersona - Prepare persona for execution
book - Book notation utilities
isValidPipelineString - Validate pipeline string
GENERIC_PIPELINE_INTERFACE - Generic pipeline interface
getPipelineInterface - Get pipeline interface
isPipelineImplementingInterface - Check interface implementation
isPipelineInterfacesEqual - Compare pipeline interfaces
EXPECTATION_UNITS - Units for expectations
validatePipelineString - Validate pipeline string format

Pipeline Preparation

isPipelinePrepared - Check if pipeline is prepared
preparePipeline - Prepare pipeline for execution
unpreparePipeline - Unprepare pipeline

Remote Server Integration

identificationToPromptbookToken - Convert ID to token
promptbookTokenToIdentification - Convert token to ID

Knowledge Scraping

_BoilerplateScraperMetadataRegistration - Boilerplate scraper registration
prepareKnowledgePieces - Prepare knowledge pieces
$scrapersMetadataRegister - Scrapers metadata registry
$scrapersRegister - Scrapers registry
makeKnowledgeSourceHandler - Create knowledge source handler
promptbookFetch - Fetch with promptbook context
_LegacyDocumentScraperMetadataRegistration - Legacy document scraper
_DocumentScraperMetadataRegistration - Document scraper registration
_MarkdownScraperMetadataRegistration - Markdown scraper registration
_MarkitdownScraperMetadataRegistration - Markitdown scraper registration
_PdfScraperMetadataRegistration - PDF scraper registration
_WebsiteScraperMetadataRegistration - Website scraper registration

Storage Backends

BlackholeStorage - Blackhole storage (discards data)
MemoryStorage - In-memory storage
PrefixStorage - Prefixed storage wrapper

Type Definitions

MODEL_VARIANTS - Available model variants
NonTaskSectionTypes - Non-task section types
SectionTypes - All section types
TaskTypes - Task types

Server Configuration

REMOTE_SERVER_URLS - Remote server URLs

💡 This package does not make sense on its own, look at all promptbook packages or just install all by npm i ptbk

Rest of the documentation is common for entire promptbook ecosystem:

📖 The Book Whitepaper

Promptbook lets you create persistent AI agents that work on real goals for your company. The Agents Server is the heart of the project — a place where your AI agents live, remember context, collaborate in teams, and get things done.

Nowadays, the biggest challenge for most business applications isn't the raw capabilities of AI models. Large language models such as GPT-5.2 and Claude-4.5 are incredibly capable.

The main challenge lies in managing the context, providing rules and knowledge, and narrowing the personality.

In Promptbook, you define your agents using simple Books — a human-readable language that is explicit, easy to understand and write, reliable, and highly portable. You then deploy them to the Agents Server, where they run persistently and work toward their goals.

Paul Smith PERSONA You are a company lawyer. Your job is to provide legal advice and support to the company and its employees. GOAL Respond to incoming legal inquiries via email and keep the company website updated with the latest legal policies. RULE You are knowledgeable, professional, and detail-oriented. KNOWLEDGE https://company.com/company-policies.pdf KNOWLEDGE https://company.com/internal-documents/employee-handbook.docx USE EMAIL USE BROWSER TEAM You are part of the legal team of Paul Smith & Associés, you discuss with {Emily White}, the head of the compliance department. {George Brown} is expert in corporate law and {Sophia Black} is expert in labor law.

Aspects of great AI agent

We have created a language called Book, which allows you to write AI agents in their native language and create your own AI persona. Book provides a guide to define all the traits and commitments.

You can look at it as "prompting" (or writing a system message), but decorated by commitments.

Commitments are special syntax elements that define contracts between you and the AI agent. They are transformed by Promptbook Engine into low-level parameters like which model to use, its temperature, system message, RAG index, MCP servers, and many other parameters. For some commitments (for example RULE commitment) Promptbook Engine can even create adversary agents and extra checks to enforce the rules.

`Persona` commitment

Personas define the character of your AI persona, its role, and how it should interact with users. It sets the tone and style of communication.

Paul Smith & Associés PERSONA You are a company lawyer. Your job is to provide legal advice and support to the company and its employees.

`Goal` commitment

Goals define what the agent should actively work toward. Unlike a chatbot that only responds when asked, an agent with goals takes initiative and works on tasks persistently on the Agents Server.

`Knowledge` commitment

Knowledge Commitment allows you to provide specific information, facts, or context that the AI should be aware of when responding.

This can include domain-specific knowledge, company policies, or any other relevant information.

Promptbook Engine will automatically enforce this knowledge during interactions. When the knowledge is short enough, it will be included in the prompt. When it is too long, it will be stored in vector databases and RAG retrieved when needed. But you don't need to care about it.

`Rule` commitment

Rules will enforce specific behaviors or constraints on the AI's responses. This can include ethical guidelines, communication styles, or any other rules you want the AI to follow.

Depending on rule strictness, Promptbook will either propagate it to the prompt or use other techniques, like adversary agent, to enforce it.

`Use` commitments

Use commitments grant the agent real capabilities — tools it can use to interact with the outside world. USE EMAIL lets the agent send emails, USE BROWSER lets it access and read web content, USE SEARCH ENGINE lets it search the web, and many more.

These are what turn a chatbot into a persistent agent that actually does work.

Paul Smith & Associés PERSONA You are a company lawyer. Your job is to provide legal advice and support to the company and its employees. GOAL Respond to incoming legal inquiries via email within 24 hours. GOAL Keep the company website updated with the latest legal policies and compliance information. RULE Always ensure compliance with local laws and regulations. RULE Never provide legal advice outside your area of expertise. RULE Never provide legal advice about criminal law. KNOWLEDGE https://company.com/company-policies.pdf KNOWLEDGE https://company.com/internal-documents/employee-handbook.docx USE EMAIL USE BROWSER USE SEARCH ENGINE

`Team` commitment

Team commitment allows you to define the team structure and advisory fellow members the AI can consult with. This allows the AI to simulate collaboration and consultation with other experts, enhancing the quality of its responses.

Promptbook Ecosystem

Promptbook is an ecosystem of tools centered around the Agents Server — a production-ready platform for running persistent AI agents.

Agents Server

The Agents Server is the primary way to use Promptbook. It is a web application where your AI agents live and work. You can create agents, give them knowledge and rules using the Book language, organize them into teams, and let them work on goals persistently. The Agents Server provides a UI for managing agents, an API for integrating them into your applications, and can be self-hosted via Docker or deployed on Vercel.

Promptbook Engine

The Promptbook Engine is the open-source core that powers everything. It parses the Book language, applies commitments, manages LLM provider integrations, and executes agents. The Agents Server is built on top of the Engine. If you need to embed agent capabilities directly into your own application, you can use the Engine as a standalone TypeScript/JavaScript library via NPM packages.

💜 The Promptbook Project

Promptbook project is an ecosystem centered around the Agents Server — a platform for creating, deploying, and running persistent AI agents. Following is a list of the most important pieces of the project:

🌐 Community & Social Media

Join our growing community of developers and users:

🖼️ Product & Brand Channels

Promptbook.studio

📚 Documentation

See detailed guides and API reference in the docs or online.

🔒 Security

For information on reporting security vulnerabilities, see our Security Policy.

📦 Deployment & Packages

The fastest way to get started is with the Agents Server:

🐋 Docker image — Self-host the Agents Server with full control over your data
☁️ Hosted Agents Server — Start creating agents immediately, no setup required

NPM Packages (for developers embedding the Engine)

If you want to embed the Promptbook Engine directly into your application, the library is divided into several packages published from a single monorepo. You can install all of them at once:

npm i ptbk

Or you can install them separately:

⭐ Marked packages are worth to try first

⭐ ptbk - Bundle of all packages, when you want to install everything and you don't care about the size
promptbook - Same as ptbk
⭐🧙‍♂️ @promptbook/wizard - Wizard to just run the books in node without any struggle
@promptbook/core - Core of the library, it contains the main logic for promptbooks
@promptbook/node - Core of the library for Node.js environment
@promptbook/browser - Core of the library for browser environment
⭐ @promptbook/utils - Utility functions used in the library but also useful for individual use in preprocessing and postprocessing LLM inputs and outputs
@promptbook/markdown-utils - Utility functions used for processing markdown
@promptbook/javascript - Execution tools for javascript inside promptbooks
@promptbook/openai - Execution tools for OpenAI API, wrapper around OpenAI SDK
@promptbook/anthropic-claude - Execution tools for Anthropic Claude API, wrapper around Anthropic Claude SDK
@promptbook/vercel - Adapter for Vercel functionalities
@promptbook/google - Integration with Google's Gemini API
@promptbook/deepseek - Integration with DeepSeek API
@promptbook/ollama - Integration with Ollama API
@promptbook/azure-openai - Execution tools for Azure OpenAI API
@promptbook/fake-llm - Mocked execution tools for testing the library and saving the tokens
@promptbook/remote-client - Remote client for remote execution of promptbooks
@promptbook/remote-server - Remote server for remote execution of promptbooks
@promptbook/pdf - Read knowledge from .pdf documents
@promptbook/documents - Integration of Markitdown by Microsoft
@promptbook/documents - Read knowledge from documents like .docx, .odt,…
@promptbook/legacy-documents - Read knowledge from legacy documents like .doc, .rtf,…
@promptbook/website-crawler - Crawl knowledge from the web
@promptbook/editable - Editable book as native javascript object with imperative object API
@promptbook/templates - Useful templates and examples of books which can be used as a starting point
@promptbook/types - Just typescript types used in the library
@promptbook/color - Color manipulation library
@promptbook/cli - Command line interface utilities for promptbooks

📚 Dictionary

The following glossary is used to clarify certain concepts:

General LLM / AI terms

Prompt drift is a phenomenon where the AI model starts to generate outputs that are not aligned with the original prompt. This can happen due to the model's training data, the prompt's wording, or the model's architecture.
Pipeline, workflow scenario or chain is a sequence of tasks that are executed in a specific order. In the context of AI, a pipeline can refer to a sequence of AI models that are used to process data.
Fine-tuning is a process where a pre-trained AI model is further trained on a specific dataset to improve its performance on a specific task.
Zero-shot learning is a machine learning paradigm where a model is trained to perform a task without any labeled examples. Instead, the model is provided with a description of the task and is expected to generate the correct output.
Few-shot learning is a machine learning paradigm where a model is trained to perform a task with only a few labeled examples. This is in contrast to traditional machine learning, where models are trained on large datasets.
Meta-learning is a machine learning paradigm where a model is trained on a variety of tasks and is able to learn new tasks with minimal additional training. This is achieved by learning a set of meta-parameters that can be quickly adapted to new tasks.
Retrieval-augmented generation is a machine learning paradigm where a model generates text by retrieving relevant information from a large database of text. This approach combines the benefits of generative models and retrieval models.
Longtail refers to non-common or rare events, items, or entities that are not well-represented in the training data of machine learning models. Longtail items are often challenging for models to predict accurately.

Note: This section is not a complete dictionary, more list of general AI / LLM terms that has connection with Promptbook

💯 Core concepts

📚 Collection of pipelines
📯 Pipeline
🙇‍♂️ Tasks and pipeline sections
🤼 Personas
⭕ Parameters
🚀 Pipeline execution
🧪 Expectations - Define what outputs should look like and how they're validated
✂️ Postprocessing - How outputs are refined after generation
🔣 Words not tokens - The human-friendly way to think about text generation
☯ Separation of concerns - How Book language organizes different aspects of AI workflows

Advanced concepts

� Agents Server

The Agents Server is the primary way to use Promptbook. It is a production-ready platform where you create, deploy, and manage persistent AI agents that work toward goals. Agents remember context across conversations, collaborate in teams, and follow the rules and knowledge you define in the Book language.

Hosted at gallery.ptbk.io — start creating agents immediately
Self-hosted via Docker — full control over your data and infrastructure
API for integrating agents into your own applications

🚂 Promptbook Engine

The Engine is the open-source core that powers the Agents Server. If you need to embed agent capabilities directly into your TypeScript/JavaScript application, you can use it as a standalone library.

Schema of Promptbook Engine

➕➖ When to use Promptbook?

➕ When to use

When you want to deploy persistent AI agents that work on goals for your company
When you need agents with specific personalities, knowledge, and rules tailored to your business
When you want agents that collaborate in teams and consult each other
When you need to integrate AI agents into your existing applications via API
When you want to self-host your AI agents with full control over data and infrastructure
When you are writing an app that generates complex things via LLM — like websites, articles, presentations, code, stories, songs,...
When you want to version your agent definitions and test multiple versions
When you want to log agent execution and backtrace issues

➖ When not to use

When a single simple prompt already works fine for your job
When OpenAI Assistant (GPTs) is enough for you
When you need streaming (this may be implemented in the future, see discussion)
When you need to use something other than JavaScript or TypeScript (other languages are on the way, see the discussion)
When your main focus is on something other than text — like images, audio, video, spreadsheets (other media types may be added in the future, see discussion)
When you need to use recursion (see the discussion)