Skip to content

Welcome to Agentic Foundry

Agentic Foundry is an open-source framework designed to empower developers to create, configure, and deploy customizable AI agents. By using predefined templates, this platform ensures seamless tool integration and streamlines agentic workflow development.

What is Agentic Foundry

The Agentic Foundry enables the efficient creation and deployment of AI agents by offering customizable templates and a set of powerful tools. These agents can be tailored to specific roles and personas, and developers can integrate their own custom tools to suit various needs.

Key Components:

1. Agents

Agents are the core building blocks of Agentic Foundry. An agent is an autonomous AI entity configured with a specific role, persona, and set of instructions that define how it reasons, plans, and acts. Agents receive user queries, determine the best course of action, invoke tools or external services, and return structured responses.

The platform provides below agent templates, each designed for a different workflow pattern:

Template Description
React Single-task reasoning with step-by-step tool execution
React Critic Adds self-critique to React for improved output quality
Multi Agent Collaborative planning and execution across specialized agents
Planner Executor Separates planning from execution with replanning support
Meta Orchestrator that dynamically coordinates worker agents
Planner Meta Advanced orchestrator with multi-prompt planning and delegation
Hybrid Combines reasoning and collaborative strategies adaptively

Agent Design Agent Configuration Inference

Supporting Components

Each agent can be equipped with the following components to extend its capabilities:

Reusable code functions that give agents the ability to perform actions — call APIs, query databases, send emails, and more. Tools are onboarded through the platform with built-in validation and testing to ensure quality, security, and reliability before deployment.

Tools Configuration

Custom validation logic — similar to tools — mapped to agents during onboarding. When enabled, validators automatically check agent responses against expected patterns, returning a validation score, status, and feedback to improve response quality in real time.

Validators

External tool and service connections via the Model Context Protocol. Register local, remote, or file-based MCP servers for real-time tool discovery, enterprise security, approval workflows, and audit logging.

MCP Registry

Domain-specific document repositories (PDF, TXT) that agents reference during inference. Upload documents from the Tool page and select one or more knowledge bases in the chat interface for accurate, context-aware answers.

Knowledge Base

2. Agent Workflows

Agent Workflows allow you to orchestrate multiple agents into complex, multi-step workflows using a visual drag-and-drop builder. Instead of manually coordinating agent calls, you define a workflow that automatically routes data between agents based on your configured flow — with conditional branching, input/output management, and reusable configurations.

Agent Workflows

How Agentic AI Differs from Traditional Software and GenAI

agentic ai

Traditional Systems:

Traditional softwares including process automation, supervised machine learning (ML), and even generative AI lacks true autonomy. These systems perform only the tasks they are explicitly programmed or trained to do. They follow predefined instructions or generate outputs within defined parameters.

What Makes Agentic AI Different:

Agentic AI is designed to operate with autonomy. Rather than executing predefined tasks, it receives high-level goals or events as input and independently determines how to act in order to achieve the desired outcomes.

Core Mechanisms of Agentic AI:

Agentic AI integrates key capabilities to enable autonomous behavior:

  • Plan: Develops a strategy to achieve the given goal or respond to an event.
  • Act: Executes tasks and may interact with external systems or tools to carry out its strategy.
  • Reflect: Evaluates the outcomes of its actions, learns from them, and adjusts its behavior accordingly.
  • Respond: Delivers a final decision or result based on its planning and learning process.

Key Characteristics:

The output of an Agentic AI system is:

  • Goal-directed
  • Interactive
  • Context-aware

Why to Use Agentic Foundry

The Agentic Foundry offers a comprehensive solution with continually evolving new features:

1. Agent-as-a-Service: Open-source framework-based solution that allows users to quickly configure agents and tools. End-users can interact with agents via a conversational interface.

2. Vertical Agents: A suite of agents tailored for various industry personas. These agents represent specific workflows for different domains, including:

  • Finance:

    • Financial statement analysis
    • Risk assessment and management
    • Fraud prevention

  • Healthcare:

    • Medical record analysis
    • Diagnostic assistance
    • Treatment plan recommendation

  • Insurance:

    • Policy underwriting assistance
    • Claim fraud detection
    • Risk assessment for policy pricing

  • Retail:

    • Customer segmentation
    • Personalized product recommendations
    • Price optimization

  • Communication:

    • Customer churn prediction
    • Fraud detection in calls and data usage
    • Customer service automation

  • Manufacturing:

    • Quality control inspection
    • Predictive maintenance
    • Product lifecycle management

The following are sample use cases categorized by domain:

Finance
  • Customer Onboarding
  • Credit Worthiness Assessment
  • Personal Finance Advisor
Healthcare
  • Personalized care plan
  • E-Triage
  • Medical translator
Insurance
  • Personalized insurance quotes & underwriting
  • Improve conversion ratio by AI assisted AI
  • Reduce CSR call efforts
Retail
  • Customer segmentation
  • Supplier recommendation
  • Competitor Product Analysis
Travel
  • Travel Operations
  • Intelligent accommodation systems
  • Travel Inspirational Video
Manufacturing
  • Demand Forecasting
  • Production Planning and Scheduling
  • Energy Utilization Analytics

3. Horizontal Agents

  • These agents offer common functionality across industries, such as:
    • Email sending
    • File search
    • Agentic RAG (Retrieval-Augmented Generation).
    • SDLC

4. Platform Capabilities

The framework goes beyond agent creation with a rich set of built-in platform features:

  • Seven Agent Templates: React, React Critic, Multi Agent, Planner Executor, Meta, Planner Meta, and Hybrid — each designed for different workflow patterns.
  • Agent Workflows: Visual drag-and-drop builder for designing multi-agent workflows with conditional branching and reusable configurations.
  • MCP Registry: Model Context Protocol integration for connecting agents to external tools and services with real-time discovery and enterprise security.
  • Memory Management: Semantic memory for cross-session fact storage and episodic memory for few-shot learning from past conversations.
  • Knowledge Base: Upload PDFs and text documents as domain-specific knowledge sources for inference.
  • Data Connectors: Native connectivity to PostgreSQL, SQLite, MySQL, and MongoDB.
  • Tool Validation: Automated validation checklist ensuring code quality, security, and reliability before tool onboarding.
  • Validators: Custom response validation logic (similar to tools) that agents use to score, verify, and improve their own outputs in real time.
  • Tool Interrupt: Interactive mode to review, edit, and approve tool parameters step-by-step before execution.
  • Canvas Screen: Rich visualization of tables, charts, graphs, and images directly in the chat interface.
  • Prompt Optimization: Automated prompt evolution using Pareto sampling and LLM-as-judge scoring.
  • Vault: Secure secrets management with private and public vaults for API keys, URLs, and credentials.
  • SSE Streaming: Real-time streaming of agent execution steps to the UI as they happen.
  • RBAC: Role-based access control with Admin, Developer, and User roles.
  • JWT Authentication: Secure API endpoint authentication via Bearer tokens.
  • Evaluation: LLM-as-a-judge evaluation, ground-truth benchmarking, and consistency/robustness testing.
  • Telemetry: OpenTelemetry, Arize Phoenix, and Grafana for observability, tracing, and dashboards.
  • Agent Export: Export complete agent packages for backup, migration, or redeployment — including GitHub push support.
  • Custom LLM Support: Plug in additional LLM providers (Anthropic, Mistral, Groq, and more) alongside Azure OpenAI and OpenAI.
  • GZIP Compression: Automatic compression of large API responses for faster load times and reduced bandwidth.
  • Time-To-Live (TTL): Automated lifecycle management for memory records, tools, agents, and MCP servers with recycle bin support.

How to Create and Use Agents

Follow these steps to set up your environment, create agents, and start using them:

Step 1 — Set Up Secrets in Vault
Store your API keys, database credentials, and service URLs — either in a private vault (personal) or a public vault (shared). Tools retrieve secrets by reference, so nothing is hardcoded.
Vault
Step 2 — Upload the Required Files
Upload database files (.db, .sqlite, .sql), PDFs, and other supporting documents the agent will need during inference.
Files
Step 3 — Connect Data Sources
Connect to external databases (PostgreSQL, SQLite, MySQL, MongoDB) that agents can query during inference.
Data Connectors
Step 4 — Create Knowledge Bases
Upload PDFs or text documents as domain-specific knowledge sources for context-aware answers.
Knowledge Base
Step 5 — Create and Onboard Tools
Develop the required tools and onboard them through the platform. Every tool passes an automated validation checklist before onboarding.
Tools Configuration
Step 6 — Create Validators (Optional)
Build custom validators that check agent responses against expected patterns. Map them to agents and enable the Validator toggle in chat.
Validators
Step 7 — Register MCP Servers (Optional)
Register MCP servers (local, remote, or file-based) in the MCP Registry for real-time tool discovery.
MCP Registry
Step 8 — Create the Agent
Build your AI agent by selecting a template (React, React Critic, Multi Agent, Planner Executor, Meta, Planner Meta, or Hybrid), adding tools, and configuring its role and persona.
Agent Configuration
Step 9 — Build Agent Workflows (Optional)
For multi-agent workflows, use the visual builder to connect agents with conditional branching, input/output management, and reusable configurations.
Agent Workflows
Step 10 — Start Using the Agent
Select your agent in the Inference section and start querying. Use Canvas Screen for visualizations, Tool Interrupt for step-by-step control, and Knowledge Base selection for contextual answers.
Inference
Step 11 — Evaluate and Monitor
Benchmark agent performance with LLM-as-a-judge, ground-truth comparison, or consistency testing. Monitor system health with Telemetry via OpenTelemetry, Arize Phoenix, and Grafana.
Evaluation · Telemetry

With Agentic Foundry, developers have the power to create highly customizable AI agents for various applications — empowering businesses and individuals to automate workflows and achieve efficiency with ease.

Quick Navigation

  • Features

    Explore all platform capabilities at a glance — from agent templates to evaluation

    Explore Features

  • Installation

    Set up Agentic Foundry on Windows, Linux, or Azure Kubernetes Service

    Installation Guide

  • Architecture

    Understand the system design, component layers, and data flow

    Architecture Overview

  • Agent Design

    Learn the design patterns behind all seven agent templates

    Agent Design

  • Tools & Validators

    Create, validate, and onboard tools and response validators

    Tools Configuration

  • Agent Workflows

    Visually build multi-agent workflows with drag-and-drop

    Agent Workflows

  • Inference

    Run agents with Canvas Screen, Tool Interrupt, and Knowledge Base selection

    Inference Overview

  • MCP Registry

    Connect agents to external tools via the Model Context Protocol

    MCP Registry

  • Memory Management

    Semantic and episodic memory for persistent, context-aware agents

    Memory

  • Security & Access

    RBAC, JWT authentication, and Vault secrets management

    Vault

  • Evaluation

    LLM-as-a-judge, ground-truth benchmarking, and consistency testing

    Evaluation

  • Telemetry

    OpenTelemetry, Arize Phoenix, and Grafana for full observability

    Telemetry