# AgentDock Documentation LLMs.txt


- [Agent Templates](https://hub.agentdock.ai/docs/agent-templates): Agent Templates are the core configuration mechanism in AgentDock, allowing you to define an agent's identity, capabilities, and behavior in a declarative JSON format.

- [AgentDock Pro](https://hub.agentdock.ai/docs/agentdock-pro): AgentDock Pro extends our open-source framework with enterprise-grade features, simplified agent creation, and powerful workflow orchestration.

- [Analytics Integration using PostHog](https://hub.agentdock.ai/docs/analytics): This document outlines how analytics, specifically using PostHog, is integrated into the AgentDock Starter Kit. The primary goal is to gather usage data when deployed (e.g., on AgentDock Hub) to understand application performance and user interaction patterns, while ensuring compatibility with core features like Vercel AI SDK tool calling.

- [Adding a New LLM Provider](https://hub.agentdock.ai/docs/adding-new-provider): This guide explains how to add a new LLM provider to the AgentDock Core framework.

- [Agent Node (`AgentNode`)](https://hub.agentdock.ai/docs/agent-node): The `AgentNode` class (`agentdock-core/src/nodes/agent-node.ts`) is the primary orchestrator for conversational agent interactions within AgentDock Core. It leverages the Vercel AI SDK for efficient streaming and multi-step tool execution.

- [AgentDock: Build Anything with AI Agents](https://hub.agentdock.ai/docs/agentdock-overview): AgentDock is an open-source, backend-first framework for building and deploying sophisticated AI agents. It's designed to be framework-agnostic and provider-independent, giving you complete control over your agent's implementation.

- [Core Architecture Overview](https://hub.agentdock.ai/docs/overview): **AgentDock Core** provides the foundational libraries and systems for building stateful, tool-using AI agents. It emphasizes modularity, type safety, and extensibility.

- [Request Flow in AgentDock Core](https://hub.agentdock.ai/docs/request-flow): This document outlines the typical sequence of events when a request is processed by AgentDock Core.

- [Response Streaming in AgentDock](https://hub.agentdock.ai/docs/response-streaming): AgentDock's response streaming system extends the Vercel AI SDK to provide enhanced functionality for orchestration, error handling, and state management.

- [Technology Stack](https://hub.agentdock.ai/docs/technology-stack): This document outlines the core technologies used in AgentDock Core and its reference implementation.

- [Model Architecture](https://hub.agentdock.ai/docs/model-system): This document explains the model architecture in the AgentDock reference implementation.

- [Conditional Transitions](https://hub.agentdock.ai/docs/conditional-transitions): This document explains how AgentDock manages transitions between different orchestration steps based on defined conditions.

- [LLM Orchestration Service](https://hub.agentdock.ai/docs/llm-orchestration): The `LLMOrchestrationService` serves as a bridge between the `AgentNode` and `CoreLLM`, adding orchestration-specific functionality to LLM interactions.

- [Orchestration Configuration](https://hub.agentdock.ai/docs/orchestration-config): This document details how to configure orchestration behavior for AgentDock agents using the agent template (`template.json` or similar).

- [Orchestration Framework Overview](https://hub.agentdock.ai/docs/orchestration-overview): Orchestration in AgentDock provides a structured way to control agent behavior, manage tool availability across different states (steps), and define sequences for complex tasks. It enables more guided, reliable, and focused agent interactions compared to allowing unrestricted tool use.

- [Orchestration Framework](https://hub.agentdock.ai/docs/): This directory contains documentation about AgentDock's orchestration framework.

- [Orchestration State Management](https://hub.agentdock.ai/docs/state-management): This document details the implementation of orchestration state management within AgentDock, focusing on how session-specific state is handled for controlling agent behavior and tool usage.

- [Step Sequencing](https://hub.agentdock.ai/docs/step-sequencing): This document explains how AgentDock enforces tool sequences within orchestration steps, ensuring tools are used in the intended order for structured tasks.

- [Provider-Agnostic API Architecture](https://hub.agentdock.ai/docs/provider-agnostic-api): This document explains how AgentDock works with multiple LLM providers through a unified API layer.

- [AgentDock Architecture](https://hub.agentdock.ai/docs/): This section provides an overview of the architecture of AgentDock Core, the foundation library that powers all AgentDock functionality.

- [NextJS Session Integration](https://hub.agentdock.ai/docs/nextjs-integration): This document explains how AgentDock's session management integrates with Next.js applications, focusing on API routes, client-side handling, and runtime considerations.

- [Session Management](https://hub.agentdock.ai/docs/): This directory contains documentation about AgentDock's session management system.

- [Session Management](https://hub.agentdock.ai/docs/session-management): Sessions in AgentDock provide the foundation for stateful, continuous interactions between users and AI agents. This document covers the core concepts, architecture, implementation details, and optimization strategies for the session management system.

- [Session Management Overview](https://hub.agentdock.ai/docs/session-overview): Sessions in AgentDock provide a foundation for stateful interactions between users and AI agents. This document outlines the core concepts, architecture, and design principles of the session management system.

- [AgentDock Core Error Handling](https://hub.agentdock.ai/docs/core-implementation): This document describes the error handling architecture in the agentdock-core library.

- [LLM Error Handling](https://hub.agentdock.ai/docs/llm-errors): How we handle errors from LLM providers and present them to users.

- [Error Handling in AgentDock](https://hub.agentdock.ai/docs/overview): This document outlines our approach to error handling in the AgentDock application using React ErrorBoundary components and specialized error handlers.

- [Creating Custom Evaluators](https://hub.agentdock.ai/docs/custom-evaluators): The AgentDock Evaluation Framework is designed for extensibility. While the built-in evaluators cover many common use cases, you will inevitably encounter scenarios requiring bespoke evaluation logic specific to your agent's tasks, data, or business rules. A core design philosophy is that a framework's true power lies in its adaptability.

- [Keyword Coverage Evaluator](https://hub.agentdock.ai/docs/keyword-coverage): The `KeywordCoverageEvaluator` checks for the presence, frequency, or coverage of specified keywords or phrases within a given text. This is a straightforward but highly effective way to ensure that essential information is included in an agent's response, or conversely, that undesirable terms are absent. Experience finds this useful for quick checks on information inclusion or policy adherence.

- [Lexical Evaluator Suite](https://hub.agentdock.ai/docs/lexical-evaluators): The AgentDock Evaluation Framework includes a suite of **Lexical Evaluators** designed for fast, deterministic, and cost-effective analysis of textual content. These evaluators operate directly on the text of agent responses or other inputs, without relying on complex NLP models or LLMs. Practical experience shows that these kinds of checks are invaluable for quick feedback loops and for validating basic textual properties before engaging more resource-intensive evaluations.

- [Lexical Similarity Evaluator](https://hub.agentdock.ai/docs/lexical-similarity): The `LexicalSimilarityEvaluator` compares two strings and calculates a score representing their textual similarity. This is distinct from semantic similarity (handled by `NLPAccuracyEvaluator`) as it focuses on the character-level or token-level makeup of the strings. Experience suggests this is useful for cases where specific phrasing or structure is expected, but minor variations are tolerable, or for comparing against known textual patterns.

- [LLM-as-Judge Evaluator](https://hub.agentdock.ai/docs/llm-judge): The `LLMJudgeEvaluator` leverages the power of a large language model (LLM) to perform nuanced, qualitative assessments of agent outputs. Unlike rule-based systems that check for specific, deterministic patterns, an LLM judge can evaluate responses based on broader criteria like coherence, relevance, helpfulness, safety, or adherence to complex instructions. Practical experience shows that for many sophisticated agent behaviors, an LLM judge is an indispensable tool for capturing the subtleties of quality.

- [NLP Accuracy Evaluator](https://hub.agentdock.ai/docs/nlp-accuracy): The `NLPAccuracyEvaluator` measures the semantic similarity between an agent's response and a provided ground truth. This is crucial for tasks where the exact wording isn't as important as the meaning or intent conveyed. Experience indicates that for question answering, summarization, or any task requiring semantic understanding, this kind of evaluation is far more insightful than simple lexical matches.

- [Rule-Based Evaluator](https://hub.agentdock.ai/docs/rule-based): The `RuleBasedEvaluator` provides essential, low-cost checks for agent outputs. Its purpose is straightforward: enforce deterministic constraints and basic validations without incurring the latency or cost of LLM calls. Experience shows that establishing these kinds of guardrails early is fundamental for building any semblance of predictable agent behavior.

- [Sentiment Evaluator](https://hub.agentdock.ai/docs/sentiment): The `SentimentEvaluator` analyzes the emotional tone of a given text, typically classifying it as positive, negative, or neutral. It can also provide a numerical score indicating the intensity of the sentiment. This is useful for ensuring agents maintain an appropriate tone or for flagging overly negative or positive responses. Experience shows this is a good first-pass check for agent demeanor.

- [Tool Usage Evaluator](https://hub.agentdock.ai/docs/tool-usage): The `ToolUsageEvaluator` is designed to assess the correctness of an agent's tool invocations. In modern agent systems, the ability to reliably and accurately use tools is paramount. This evaluator checks if the agent called the right tools, with the right arguments, and in the expected manner. Deploying agents has shown that tool use is a frequent point of failure, making robust evaluation in this area critical.

- [Toxicity Evaluator](https://hub.agentdock.ai/docs/toxicity): The `ToxicityEvaluator` scans text for the presence of predefined toxic terms, offensive language, or other undesirable content based on a blocklist. This is a fundamental safety check to help ensure agents do not produce harmful or inappropriate responses. Experience underscores that even basic blocklist checks are a necessary first line of defense for responsible agent deployment.

- [AgentDock Evaluation Framework: Measuring What Matters](https://hub.agentdock.ai/docs/): The capability to build AI agents is rapidly becoming commoditized. The real differentiator lies in the ability to systematically and reliably measure agent quality. Without robust evaluation, "improvement" is guesswork, and "reliability" is a marketing slogan. Experience in deploying these systems has consistently shown that what isn't measured, isn't managed, and certainly isn't improved in a way that stands up to real-world demands.

- [Getting Started with AgentDock](https://hub.agentdock.ai/docs/getting-started): This guide will help you set up and run AgentDock on your local machine for development and testing purposes.

- [AgentDock: ابنِ أي مشروع باستخدام وكلاء الذكاء الاصطناعي](https://hub.agentdock.ai/docs/): <p align="center">

- [AgentDock: 用 AI Agent 构建无限可能](https://hub.agentdock.ai/docs/): <p align="center">

- [AgentDock: Erschaffe grenzenlose Möglichkeiten mit KI-Agenten](https://hub.agentdock.ai/docs/): <p align="center">

- [AgentDock: Onbegrensde mogelijkheden met AI-Agenten](https://hub.agentdock.ai/docs/): <p align="center">

- [AgentDock : Créez sans limites avec des Agents IA](https://hub.agentdock.ai/docs/): <p align="center">

- [AgentDock: Δημιουργήστε Απεριόριστες Δυνατότητες με Πράκτορες AI](https://hub.agentdock.ai/docs/): <p align="center">

- [AgentDock: Crea Possibilità Infinite con Agenti AI](https://hub.agentdock.ai/docs/): <p align="center">

- [AgentDock: AIエージェントで無限の可能性を創造](https://hub.agentdock.ai/docs/): <p align="center">

- [AgentDock: AI 에이전트로 무한한 가능성을 창조하세요](https://hub.agentdock.ai/docs/): <p align="center">

- [AgentDock: Twórz Nieograniczone Możliwości z Agentami AI](https://hub.agentdock.ai/docs/): <p align="center">

- [AgentDock README Translations](https://hub.agentdock.ai/docs/): Welcome to the translated versions of the AgentDock README.

- [AgentDock: Создавайте Безграничные Возможности с Помощью ИИ-Агентов](https://hub.agentdock.ai/docs/): <p align="center">

- [AgentDock: Crea Posibilidades Ilimitadas con Agentes de IA](https://hub.agentdock.ai/docs/): <p align="center">

- [AgentDock: Yapay Zeka Ajanları ile Sınırsız Olasılıklar Yaratın](https://hub.agentdock.ai/docs/): <p align="center">

- [AgentDock: Створюйте Безмежні Можливості з Агентами ШІ](https://hub.agentdock.ai/docs/): <p align="center">

- [Initial Security Audit](https://hub.agentdock.ai/docs/initial-security-audit): This document summarizes the findings of an initial security review performed on the AgentDock codebase (focusing on `agentdock-core` and the open source client layer). It identifies potential vulnerabilities, analyzes risks, and tracks actions taken or recommended.

- [AgentDock Memory Architecture Overview](https://hub.agentdock.ai/docs/architecture-overview): ```mermaid

- [Complete AgentDock Configuration Guide](https://hub.agentdock.ai/docs/complete-configuration-guide): ```bash

- [Memory Consolidation Guide](https://hub.agentdock.ai/docs/consolidation-guide): Memory consolidation in AgentDock is an intelligent optimization system that prevents memory bloat while improving knowledge quality through automatic conversion, deduplication, and synthesis.

- [Memory Graph Architecture](https://hub.agentdock.ai/docs/graph-architecture): > **Technical documentation for developers and architects**

- [Memory Connections](https://hub.agentdock.ai/docs/memory-connections): AgentDock's memory connection system automatically discovers relationships between memories, enabling agents to understand context and provide more relevant responses. The system uses cost-optimized smart triage for 65% cost reduction.

- [AgentDock Memory System](https://hub.agentdock.ai/docs/): > A four-layer memory architecture that gives AI agents human-like memory capabilities

- [Research Foundations of Memory Connections](https://hub.agentdock.ai/docs/research-foundations): > **Scientific basis for AgentDock's memory connection system**

- [AgentDock: Conversational Retrieval-Augmented Generation (RAG)](https://hub.agentdock.ai/docs/retrieval-augmented-generation): > The AgentDock Recall Service implements **Conversational RAG** (Retrieval-Augmented Generation) specifically optimized for agent memory and real-time knowledge augmentation.

- [Custom Node Development](https://hub.agentdock.ai/docs/custom-node-development): This guide explains how to create custom nodes for AgentDock Core.

- [Custom Tool Development](https://hub.agentdock.ai/docs/custom-tool-development): This guide provides a comprehensive overview of creating custom tools for AgentDock, with complete examples and best practices.

- [Node System](https://hub.agentdock.ai/docs/): AgentDock is built around a powerful **node-based architecture** using `BaseNode` as the foundation for all functionality. This design allows for modular, extensible, and highly configurable systems. `BaseNode` supports various node categories, including core functionalities like LLM interaction (`AgentNode`) and callable tools, forming the basis for sophisticated agent behaviors.

- [Bring Your Own Keys (BYOK) Mode](https://hub.agentdock.ai/docs/byok-mode): BYOK (Bring Your Own Keys) mode is a security setting in the AgentDock Open Source Client that controls how API keys are managed. When enabled, AgentDock will **only** use API keys that have been explicitly provided by the user through the settings interface, and will never fall back to environment variables.

- [Diagram Examples in the Open Source Client](https://hub.agentdock.ai/docs/diagram-example): This page demonstrates how to create and render various types of diagrams using Mermaid in the AgentDock Open Source Client. These diagram examples can be used for visualizing different aspects of your application architecture, workflows, and components when building with AgentDock.

- [Image Generation](https://hub.agentdock.ai/docs/image-generation): The AgentDock Open Source Client includes a dedicated image generation feature that demonstrates how to integrate advanced AI capabilities into applications built with AgentDock Core.

- [Open Source Client (Next.js Implementation)](https://hub.agentdock.ai/docs/nextjs-implementation): This reference implementation, built with Next.js and the App Router, serves as a practical example of how to consume and interact with the AgentDock Core framework to build a full-featured web application for conversational AI agents.

- [PRD: AgentDock Evaluation Framework - Building for Measurable Quality](https://hub.agentdock.ai/docs/evaluation-framework): As AgentDock evolves, the ability to systematically measure and improve agent quality becomes paramount. While various ad-hoc methods and external tools have been used previously, a standardized, integrated **Evaluation Framework** within AgentDock Core is essential to:

- [Memory Evolution Tracker: Complete Observability for Intelligent Memory Systems](https://hub.agentdock.ai/docs/memory-evolution-tracker): > Transform your memory system from a black box into a transparent, analyzable intelligence layer

- [Memory & Storage Testing System PRD](https://hub.agentdock.ai/docs/memory-storage-testing-prd): **Product Requirements Document**

- [Workflow Learning Service - PRD](https://hub.agentdock.ai/docs/workflow-learning-service-prd): **Author**: AgentDock Team

- [AgentDock: Build Anything with AI Agents](https://hub.agentdock.ai/docs/): [Français](/docs/i18n/french/README.md) • [日本語](/docs/i18n/japanese/README.md) • [한국어](/docs/i18n/korean/README.md) • [中文](/docs/i18n/chinese/README.md) • [Español](/docs/i18n/spanish/README.md) • [Italiano](/docs/i18n/italian/README.md) • [Nederlands](/docs/i18n/dutch/README.md) • [Deutsch](/docs/i18n/deutsch/README.md) • [Polski](/docs/i18n/polish/README.md) • [Türkçe](/docs/i18n/turkish/README.md) • [Українська](/docs/i18n/ukrainian/README.md) • [Ελληνικά](/docs/i18n/greek/README.md) • [Русский](/docs/i18n/russian/README.md) • [العربية](/docs/i18n/arabic/README.md)

- [Contributing Community Agents](https://hub.agentdock.ai/docs/add-agent): AgentDock welcomes community contributions for new and interesting agents! This guide explains how to add your agent to the public `/agents` directory so it can be potentially featured in the Open Source Client and shared with others.

- [RFA-001: AI Code Reviewer Agent](https://hub.agentdock.ai/docs/001-code-reviewer): Manual code reviews are essential for maintaining code quality, but they are often time-consuming, inconsistent, and prone to human error. Reviewers can miss subtle bugs or security vulnerabilities, and feedback styles can vary widely. Developers spend significant time waiting for reviews or performing repetitive checks, slowing down the development cycle.

- [Request For Agents (RFA)](https://hub.agentdock.ai/docs/): > Build powerful AI agents with AgentDock's node-based architecture

- [Roadmap](https://hub.agentdock.ai/docs/roadmap): This document outlines the planned features and future direction for AgentDock. Most improvements target the core AgentDock framework (`agentdock-core`), which is under active development and will be published as a versioned NPM package upon reaching a stable release. Some items may also involve the open-source client.

- [Agent Marketplace](https://hub.agentdock.ai/docs/agent-marketplace): The Agent Marketplace enables creators to publish, distribute, and monetize their AI agent templates and workflows, while helping users discover high-quality solutions tailored to their needs.

- [Code Playground](https://hub.agentdock.ai/docs/code-playground): The Code Playground enables AI agents to generate, execute, and visualize code directly within the chat interface, providing a seamless environment for coding, debugging, and demonstration.

- [Agent Evaluation Framework](https://hub.agentdock.ai/docs/evaluation-framework): The Agent Evaluation Framework provides tools for measuring and improving agent performance, ensuring consistent quality across different use cases.

- [HTTP Adapter Framework Abstraction](https://hub.agentdock.ai/docs/http-adapter-framework-abstraction): **Status**: Critical Prerequisite

- [Model Context Protocol (MCP) Integration](https://hub.agentdock.ai/docs/mcp-integration): AgentDock Core will integrate the Model Context Protocol (MCP) to enable standardized tool discovery and execution across different AI agents and services. This integration will allow AgentDock agents to leverage external tools while maintaining our core architecture principles.

- [Multi-Agent Collaboration](https://hub.agentdock.ai/docs/multi-agent-collaboration): **Status: Planned**

- [Natural Language AI Agent Builder](https://hub.agentdock.ai/docs/nl-agent-builder): The Natural Language AI Agent Builder allows users to create sophisticated AI agents and workflows using plain language descriptions instead of complex coding or visual programming.

- [Platform Integration](https://hub.agentdock.ai/docs/platform-integration): The Platform Integration feature extends AgentDock agents to interact with users through external messaging platforms such as Telegram, WhatsApp, and Slack.

- [🗺️ AgentDock Development Roadmap](https://hub.agentdock.ai/docs/): This document outlines the planned features and future direction for AgentDock. Most improvements target the core AgentDock framework (`agentdock-core`), which is under active development and will be published as a versioned NPM package upon reaching a stable release. Some items may also involve the open-source client.

- [Telemetry & Observability](https://hub.agentdock.ai/docs/telemetry): The Telemetry & Observability feature provides monitoring, tracing, and evaluation capabilities for AgentDock agents, enabling developers to gain insights into agent behavior and optimize performance.

- [Voice AI Agents](https://hub.agentdock.ai/docs/voice-agents): The Voice AI Agents feature enables real-time voice conversations with AgentDock agents through advanced speech-to-speech capabilities, creating natural, interactive experiences through web applications and phone systems.

- [AgentDock Workflows & Nodes: Automating Complex Tasks](https://hub.agentdock.ai/docs/workflow-nodes): AgentDock is expanding beyond conversational agents to enable the construction of sophisticated, custom automations. The goal is to allow users to orchestrate complex processes, connect diverse tools, and integrate AI logic using a structured, node-based system, reducing the need for extensive custom code for many common automation patterns.

- [Storage Setup Guide](https://hub.agentdock.ai/docs/getting-started): ```bash

- [Message History Management](https://hub.agentdock.ai/docs/message-history): This document explains how AgentDock manages conversation history to optimize context window usage while maintaining conversation coherence.

- [Message Persistence in AgentDock](https://hub.agentdock.ai/docs/message-persistence): This document outlines the current approach to message persistence in the AgentDock reference implementation.

- [Storage System](https://hub.agentdock.ai/docs/): AgentDock Core provides a comprehensive storage abstraction layer with 15 production-ready adapters for various data persistence needs. This fully-implemented system enables switching between different storage backends without changing application code.

- [Storage Abstraction](https://hub.agentdock.ai/docs/storage-abstraction): AgentDock includes a comprehensive storage abstraction layer that provides a unified interface for multiple storage backends, enabling developers to switch between different storage solutions without changing application code.

- [Vector Storage](https://hub.agentdock.ai/docs/vector-storage): Vector storage provides embedding-based retrieval for memories, documents, and semantic search across the AgentDock platform.

- [Testing Strategy](https://hub.agentdock.ai/docs/testing): This document outlines the testing strategy for the AgentDock project, with a particular focus on the `agentdock-core` framework and its comprehensive unit testing approach.

- [Token Usage Tracking & Smart Context Management](https://hub.agentdock.ai/docs/token-usage-tracking): AgentDock Core provides mechanisms for tracking LLM token usage and plans for advanced features to manage the context window intelligently.