FoundationaLLM

The operating system for AI agents.

Deploy secure, extensible AI agents inside your own cloud. Designed for scale, built for control, and ready to work.

FoundationaLLM is a governed agent platform that runs in your own cloud environment. End users work with agents through the Agent Portal. Administrators use the Management Portal, APIs, and SDKs to create agents, connect enterprise data, and control who can use them.

Start here

Area	What you will find
Architecture and concepts	How the platform is put together, and the core resources it manages
Management Portal	Create and manage agents, data, models, access, and branding
APIs and SDKs	Core API, Management API, and the .NET and Python SDKs
Deployment	Deploy FoundationaLLM in your Azure environment
Access control	Role-based access from the instance down to an individual agent
Development	Local setup, calling the APIs, and contributing
Release notes	Current release, earlier releases from 0.9.7, and breaking changes

What the platform includes

These are the capabilities described on foundationallm.ai. Use the links to go to the matching documentation.

Agents

Build an agent once and run it on Google, OpenAI, Anthropic, Mistral, and other models. Behavior is configuration-driven: conversation history, gatekeeper security, prompt rewriting, semantic caching, and real-time speech are set per agent.

Capability	Documentation
Create and manage agents	Management Portal
Real-time speech	Enabling real-time speech
Tools and MCP servers	MCP client tool
Agent memory	Personalized memory
Quotas and usage limits	Configuring quotas

The Agent Portal is the conversational interface for end users. It supports conversation management, file upload, and specialized tools. The same agent experience can be embedded in a website as a docked, floating, or full-screen widget. Agents can also run on a schedule, use approved self-service templates, keep memory across interactions, and produce multi-step research deliverables.

Data

Built-in data pipelines turn enterprise data into knowledge agents can retrieve. Shielded pipelines scan files for prompt injection before indexing.

Capability	Documentation
Data pipelines	Creating data pipelines
Pipeline concepts	Data pipeline concepts

Security and operations

Agents run in your own cloud, so data stays in your tenant. Access is role-based. Quotas keep consumption predictable. Telemetry, analytics, and abuse detection show how the platform is used. A hosted sandbox gives agents a governed place to run the software they need, including a custom code interpreter.

Capability	Documentation
Authentication	Authentication and authorization
Role-based access	Role-based access control
Sandbox	Sandbox deployment
Deployment options	Quick start and standard deployment

For developers

Call the platform directly, or use an SDK.

Surface	Documentation
Core API	Chat completions, sessions, and files
Management API	Agents, data sources, and configuration
.NET SDK	.NET SDK
Python SDK	Python SDK

Why a platform instead of a framework

FoundationaLLM is the agent runtime and control plane. Frameworks such as LangChain or Semantic Kernel are libraries you assemble yourself. The platform already includes orchestration, role-based access, audit, content safety, quotas, reusable agent configuration, and a model- and tool-agnostic architecture.

Assembling it yourself	FoundationaLLM
Orchestration spread across libraries	An agent-first orchestration layer
Governance, quotas, and access added later	Role-based access, audit, and content safety included
Each team builds its own path to production	Reusable workflows and fast agent configuration
Tied to one model or one tool set	LLM- and tool-agnostic

More about the product is on foundationallm.ai. The source is on GitHub.