The Open Agent Hub Projects
Collaborate, experiment, and build production-ready, open-source agents.
See the Open Agent Hub website.
Direct Links
- Submit a project for us to support or adopt (defined below).
- Submit a use case for your domain. Help us help you build it!
The Open Agent Hub is a collaborative community of open-source AI projects and domain-specific work groups that seek to make AI Agents successful in the real world through fast experimentation and distillation of learning into reusable reference architectures and enterprise-quality implementations.
Our focus on domain-specific projects helps surface and address the unique challenges faced in particular domains, which apply in other domains, too. This is why we need to know your important use cases!
Our work groups include engineers, AI researchers, and subject matter experts from industry-leading organizations. Here is some of our work so far:
Industrial AI: Examples include domain-specific models for Semiconductor process agents (SemiKong - paper) and Marine Navigation (Llamarine - paper).
Finance, Medical, and Other Research Areas: Deep Research Agent for Applications is a collaboration between domain and AI experts to explore the practical challenges of building and running trustworthy, production-quality deep research applications, with example applications for finance, medicine, and ArXiv exploration.
Expert Knowledge / Graphs: Semiont is a Wiki-like knowledge base supporting graph retrieval, where humans and agents co-create Knowledge. Bartlebot is a demonstration AI Agent for the legal domain with a Slack integration.
Geospatial: GeoBench is a general Earth observation benchmark for evaluating the performance of large models on geospatial data, and TerraTorch is a Python toolkit for fine-tuning Geospatial Foundation Models (GFMs).
Chemistry and Materials: Foundation models for molecular analysis.
We welcome your feedback and help, including suggestions for new projects and domain-specific use cases of importance to you.
Agent Ecosystems with MCP, A2A, and Related Projects
The Model Context Protocol (MCP) from Anthropic is quickly becoming an industry standard for communications between models, tool, and data repositories. A competing project with more emphasis on the unique requirements for agents is the [Agent2Agent](https://a2a-protocol.org/latest/] (A2A){:target=”_blank”} protocol. The AI Alliance seeks to advance the application of these protocols and related tools to foster the development of robust distributed AI systems.
| Links | Description |
|---|---|
MCP (and Beyond) in the Enterprise: A User Guide  1
|
|
| Model Context Protocol (MCP) has enormous potential to accelerate AI adoption in enterprises. Alternative protocols and complementary tools are also emerging rapidly. This "living" user guide features chapters written by experts on various aspects of deploying, managing, and using these tools successfully in enterprise settings. It contains the first several chapters with many more coming soon. (Contributions are welcome!.) | |
| Context Forge | |
| Context Forge is an AI application management suite, with support for protocols like MCP, A2A, REST, etc. It serves as a central management point for tools, resources, and prompts, as well as observability, security, and access control. (Principal developer: IBM) | |
|
Deep Research Agent for Applications
|
|
| The Deep Research Agent for Applications demonstrates MCP in action for an important, common design pattern, Deep Research Agents. The first example application shows how a financial analyst can use a deep research agent to find, aggregate, and analyze information about a public company (or other potential investment). The second example explores recent medical research on diseases and pharmaceuticals. The third example supports finding and summarizing recent research papers posted to [ArXiv](https://arxiv.org){:target="_blank"}. There are many other applications possible. The app is built on MCP Agent, developed by LastMile AI, discussed next. | |
| LastMile AI MCP Agent | |
| Build effective agents using Model Context Protocol and simple to sophisticated workflow patterns. See the Deep Research Agent for Applications, discussed in the previous row, which is built with this toolkit. See the recent Alliance blog post on their lessons learned developing the orchestration feature for deep research and related use cases. Highly informative! (Principal developer: LastMile AI) | |
1 The icon indicates an Alliance core project.
The NLIP Project
The NLIP project is facilitating the development of an open-source protocol for intelligent agents to communicate with each other and with humans using natural language. The MCP (and Beyond) in the Enterprise: A User Guide, discussed above has a chapter on NLIP.
| Links | Description |
|---|---|
| NLIP Project | |
|
The NLIP project is facilitating the development of an open-source protocol for intelligent agents to communicate with each other and with humans using natural language. NLIP is designed to perform the role of a meta-protocol that allows agents from other ecosystems to communicate with one another including interfaces with other protocols such as A2A, ACP, AGNTCY, MCP, NANDA, etc.
One outcome will be a new ECMA standard, TC-56 NLIP, Natural Language Interaction Protocol (draft). The organization is also developing reference implementations of the protocol and end-points. See the GitHub organization for details on these implementations. |
Agent Development Tools
See also Deep Research Agent for Applications, discussed above.
| Links | Description |
|---|---|
|
Testing Generative AI Agent Applications
|
|
| Are you an enterprise developer? How should you test AI applications? You know how to write deterministic tests for your "pre-AI" applications. What should you do when you add generative AI models, which aren't deterministic? This project adapts existing evaluation techniques for the "last mile" of AI evaluation; verifying that an AI application correctly implements its requirements and use cases, going beyond the general concerns of evaluation for safety, security, etc. We are building nontrivial, reusable examples and instructional materials, so you can use these techniques effectively in combination with the traditional tools you already know. (This project is also discussed under the AI Safety, Governance, and Education projects.) | |
|
CUBE Standard
|
|
|
Common Unified Benchmark Environment meets a common necessity, to standardize benchmark wrapping so the community can integrate otherwise-incompatible benchmarks uniformly and use them everywhere.
(Principal developer: ServiceNow) |
|
|
CUBE Harness
|
|
|
CUBE Harness is an open-source framework and research initiative for building and evaluating UI agents.
(Principal developer: ServiceNow) |
|
| Configurable Generalist Agent | |
| CUGA is an open-source generalist agent framework from IBM Research, purpose-built for enterprise automation. Designed for developers, CUGA combines and improves the best of foundational agentic patterns such as ReAct, CodeAct, and Planner-Executor into a modular architecture enabling trustworthy, policy-aware, and composable automation across web interfaces, APIs, and custom enterprise systems. | |
| Dana — The Agent-Native Evolution of AI Development | |
|
Dana is based on the question, “What if your agents could learn, adapt, and improve itself in production—without you?”
Dana bridges the gap between AI coding assistance and autonomous agents through agent-native programming: native agentprimitives, context-aware reason() calls that adapt output types automatically, self-improving pipelines with compositional | (“pipe”) operators, and functions that evolve through POET feedback loops (an automated prompt improvement technique). (Principal developer: Aitomatic)
|
|
Agent Knowledge and Tool Foundations
See also the Deep Research Agent for Applications, which is discussed above.
| Links | Description |
|---|---|
|
Gofannon
|
|
| A repository of functions consumable by other agent frameworks. | |
|
Semiont
|
|
| Wiki-like knowledge base supporting graph retrieval, where humans and agents co-create Knowledge. Includes MCP server. | |
|
Proscenium
|
|
| Collaborative, Asynchronous Human/Agent Interactions. | |
|
Lapidarist
|
|
| Document enrichment and knowledge structure (e.g., knowledge graph) extraction and resolution. | |
|
AllyCat
|
|
| (Beginner friendly!) Get started with a simple and fun end-to-end RAG application that scrapes your website so you can ask it questions. | |
|
Bartlebot
|
|
| Bartlebot is a demonstration of an AI Agent for the legal domain with a Slack integration. It is in early development. | |
|
The Living Guide to Applying AI
|
|
| Tips from experts on using AI for various applications, including popular design patterns. | |
Llama Stack and Llama Stack Agents
The Llama Stack project standardizes the core building blocks that simplify AI application development. It codifies best practices across the Llama ecosystem, integrates with other open-source tools and managed services, and provides APIs for inference, evaluation, agents, MCP, and deployment requirements like observability. It is designed to support both on-premise and cloud deployments. The ecosystem provides many example applications to help developers build and deploy AI applications quickly and effectively.
AI Alliance members are contributing directly to Llama Stack development, as well as building example applications that illustrate its use in various enterprise scenarios. The llama-stack-examples project has two initial example applications, described in the table below. The first app is a simple getting-started chatbot that shows you the basics of creating an app with Llama Stack and how to run it. The second app (in development) is a deep research application, a popular class of AI applications, which will demonstrate Llama Stack support for technologies like agents and MCP. Other examples under consideration will be chosen to cover other common application patterns seen in several industries. Please join us!.
| Links | Description |
|---|---|
| Llama Stack | |
|
The Llama Stack project itself. See also the Llama Stack Python Client. |
|
| Llama Stack Example Apps | |
|
A growing suite of example applications for Llama Stack that demonstrate how to build applications that use the RAG pattern and agents. See also the Llama Stack Demos for OpenShift and Kubernetes. |
|
|
AI Alliance Llama Stack Example Apps
|
|
A growing suite of example applications for Llama Stack that demonstrate various stack features and common application patterns:
|
|
| CCVec - Common Crawl to Vector Stores | |
| Search, analyze, and index Common Crawl data into vector stores for RAG applications, with three interfaces: CLI, Python library, and an MCP server. (Principal developers: Common Crawl Foundation and Meta) | |
| Red Hat Lightspeed | |
| An end-to-end system management tool that predicts risks across Red Hat platforms, recommends actions, and tracks costs. Red Hat Lightspeed uses AI-powered package recommendations and planning capabilities to provide targeted guidance on increasing your systems’ day-to-day efficiency. (Principal developer: Red Hat) | |
Deployment and Scaling
Deploying and scaling AI systems is critical for cost-effective use of AI. There is the growing diversity of hardware accelerators for AI, not only for servers, but for edge devices, too. Developers want the ability to write AI applications that efficiently and transparently scale across different deployment scenarios, from PoCs and single-node deployments on development laptops and edge devices, up to large-scale clustered deployments supporting many users.
| Links | Description |
|---|---|
|
The AI Accelerator Software Ecosystem Guide
|
|
| A guide to the most common AI accelerators and the software stacks they use to integrate with tools you know, like PyTorch. | |