Skip to content

Features

Agentic Foundry provides comprehensive capabilities for building and managing intelligent agents with minimal coding effort.


Features Overview

Core Development
Low-Code Agent Creation
Provide tool logic and workflow definitions — the framework handles the rest.
🛠️
Agent & Tool Management
Onboard, update, or remove agents and tools through an intuitive interface.
🎨
In-Platform Customization
Customize tools, workflows, and control logic directly — no external IDEs required.
🧩
Reusable Components
Modular agents and tools reusable across workflows and projects.
Viber Agent
Conversational AI assistant that guides you to create agents from plain descriptions — no technical knowledge needed.
Workflow & Control
🧑‍💼
Human-in-the-Loop
Manual checkpoints for human review and intervention at critical steps.
🔗
Agent Workflows
Visual drag-and-drop builder for multi-agent workflows with conditional branching.
🎯
Orchestrator Agent
Central coordinating agent that manages tasks across distributed agents.
Tool Interrupt
Review and approve tool executions step-by-step before processing.
⚙️
Dynamic Workflow Automation
Adaptive workflows with dynamic branching and conditional task handling.
📡
SSE Streaming
Real-time streaming of each agent execution step to the UI as it happens.
🖼️
Canvas Screen
Rich visualization of tables, charts, graphs, and images in chat inference.
Prompt Optimization
Automated prompt evolution using Pareto sampling and LLM-as-judge scoring.
Validators
Custom response validation logic mapped to agents — scoring, verifying, and improving outputs in real time.
Intelligence & Learning
🧠
Semantic Memory
Persistent cross-session fact storage and retrieval via Redis and PostgreSQL.
💡
Episodic Memory
Few-shot learning from past conversations using similarity scoring.
📚
Custom Knowledge Bases
Upload documents (PDF, TXT) for domain-specific agent intelligence.
🔄
Feedback-Driven Learning
Continuous improvement through structured user feedback loops.
Data & Integrations
🗄️
Data Connectors
Connect to PostgreSQL, SQLite, MySQL, and MongoDB from agent workflows.
🔌
MCP Registry
Model Context Protocol integration with real-time tool discovery and audit logging.
🤖
Flexible Model Support
Plug in Azure OpenAI, OpenAI, or custom LLM providers seamlessly.
🖥️
Model Server Integration
Centralized hosting of bi-encoder and cross-encoder models via FastAPI.
Security & Access
🛡️
Role-Based Access Control
Admin, Developer, and User roles with granular platform permissions.
🔑
JWT Authentication
Secure API authentication using Bearer tokens for all endpoints.
🔒
Vault (Secrets Management)
Private and public vaults for API keys, URLs, and credentials.
📦
Agent Export
Export complete agent packages for backup, migration, or redeployment.
Evaluation & Monitoring
⚖️
LLM-Based Evaluation
LLM-as-a-judge scoring with side-by-side model comparison.
🎯
GroundTruth Evaluation
Compare agent responses against expected outputs via CSV/XLSX upload.
📊
Consistency & Robustness
Temporal consistency and adversarial robustness testing for agents.
📈
Telemetry & Monitoring
OpenTelemetry, Arize Phoenix, and Grafana for observability and alerts.

Feature Descriptions

Core Development Features

1. Low-Code Agent Creation

Reduced development time with a low-code approach. Simply provide the tool logic and workflow definitions — the framework automatically handles the rest.

2. Agent & Tool Management

Seamlessly onboard, update, or remove agents and tools through an intuitive interface.

3. In-Platform Customization

Customize tools, workflows, and control logic directly within the framework — no external IDEs or redeployment required.

4. Reusable Components

Design agents and tools as modular, self-contained components. These can be reused across multiple workflows and projects, promoting consistency and accelerating development.


Workflow & Control Features

5. Human-in-the-Loop

Integrate manual checkpoints into automated workflows to enable human review and intervention during critical decision-making steps. Maintain oversight and control where it matters most, especially in sensitive or high-risk operations.

6. Agent Workflows

Visually design multi-agent workflows using a drag-and-drop canvas. Connect multiple agents in sequence or parallel with conditional branching, input/output management, and reusable workflow configurations.

7. Orchestrator Agent

Manage complex multi-agent systems through a central coordinating agent. The Orchestrator assigns tasks, handles inter-agent communication, resolves conflicts, and ensures that distributed agents work toward a unified goal.

8. Tool Interrupt

Review, modify, and approve tool executions before they are processed. This interactive mode provides step-by-step control over tool calls, allowing parameter editing and approval at each stage.

9. Dynamic Workflow Automation

Automate workflows that can adapt to real-time inputs, system feedback, or environmental changes. The platform supports dynamic branching, state-aware execution, and conditional task handling.

10. SSE Streaming

Server-Sent Events stream each step of the agent's internal processing to the UI in real-time — providing complete visibility into tool calls, reasoning steps, and execution progress as they happen.

11. Canvas Screen

An advanced visualization feature in chat inference that automatically renders structured data as tables, charts, graphs, or images — providing rich, context-aware output directly in the chat interface.

12. Prompt Optimization

An automated system that generates, tests, and evolves multiple prompt versions using Pareto sampling and LLM-as-judge scoring to find the most accurate and efficient system prompt for your agent.

13. Validators

Create custom validation logic — similar to tools — and map them to agents during onboarding. When enabled via the Validator toggle in chat, validators automatically check agent responses against expected patterns, returning a validation score, status, and feedback. Supported across React, React Critic, Planner Executor, Meta, and Planner Meta templates. Validators help ensure response accuracy, format compliance, and overall quality in real time.


Intelligence & Learning Features

14. Semantic Memory

Persistent cross-session memory that stores and retrieves user-provided facts, preferences, and contextual information using Redis and PostgreSQL — enabling personalized, context-aware interactions.

15. Episodic Memory

Agents learn from past conversational experiences by storing query-response examples. Using bi-encoder and cross-encoder similarity scoring, agents apply few-shot learning to improve future responses based on positive and negative feedback.

16. Custom Knowledge Bases

Equip agents with domain-specific intelligence by uploading documents (PDF, TXT) as knowledge bases. Agents reference these during inference for accurate, context-aware answers.

17. Feedback-Driven Learning

Continuously improve agent performance through direct user feedback. The system supports structured and unstructured feedback loops that fine-tune decision-making, language understanding, and tool usage over time.


Data & Integrations

18. Data Connectors

Connect to and interact with PostgreSQL, SQLite, MySQL, and MongoDB databases directly from agent workflows through a simple connection interface.

19. MCP Registry

Model Context Protocol integration for connecting agents to external tools and services. Supports local, remote, and file-based MCP servers with real-time tool discovery, enterprise security, and audit logging.

20. Flexible Model Support

Plug in different LLMs or SLMs as per task needs. Supports Azure OpenAI, OpenAI, and custom LLM providers through a straightforward configuration process.

21. Model Server Integration

Centralized hosting of bi-encoder and cross-encoder models via a FastAPI-based model server — eliminating redundant downloads and local storage across multiple environments.


Security & Access

22. Role-Based Access Control (RBAC)

Three distinct user roles — Admin, Developer, and User — each with specific access levels and permissions across the platform including tools, agents, vault, data connectors, and inference features.

23. JWT Authentication

Secure API endpoint authentication using JSON Web Tokens. All API requests are authorized via Bearer tokens, ensuring integrity and controlled access to platform endpoints.

24. Vault (Secrets Management)

A secure storage system for API keys, URLs, and credentials with private (user-only) and public (organization-wide) vaults. Tools retrieve secrets by reference — no hardcoded values required.

25. Agent Export

Export complete agent configurations including tools, dependencies, static files, validators, and SSE configurations as a self-contained package for backup, migration, or redeployment across environments.


Evaluation & Monitoring

26. LLM-Based Evaluation

Assess agent quality using the LLM-as-a-judge methodology. Compare two models side-by-side across predefined metrics with real-time progress updates via SSE.

27. GroundTruth Evaluation

Measure agent performance by comparing generated responses against expected outputs provided via CSV or XLSX upload. Produces comprehensive accuracy and quality metrics.

28. Consistency & Robustness Evaluation

Evaluate response stability across identical queries over time (consistency) and resilience against edge cases, malformed inputs, and adversarial scenarios (robustness).

29. Telemetry & Monitoring

Monitor system performance and behavior using OpenTelemetry for distributed tracing, Arize Phoenix for model observability, and Grafana for real-time dashboards and alerts.