A collection of 50+ runnable Jupyter notebook tutorials covering GenAI agent techniques, from simple chatbots to multi-agent systems.
cuga-project/cuga-agent
Enterprise agent harness that turns business requests into actions across apps, APIs and documents, with policy controls and private deployment options.
- Stars
- 1,137
- Forks
- 208
- Open issues
- 244
- Last commit
- 2 days ago
What is it?
CUGA is an open-source generalist agent harness aimed at enterprise automation. It executes multi-step tasks across web interfaces, APIs, and documents, and connects tools through MCP servers, OpenAPI services, and Python functions. Teams can configure agents and supervisors in a browser-based management UI, test a draft, and publish a version.
The project focuses on control. Execution policies can block disallowed intents, require human approval for selected tools, guide workflows with playbooks, and enforce tool-level guards before a call runs. It supports OpenAI-compatible private inference endpoints, Kubernetes deployment via a Helm chart, and an air-gapped image with preloaded supporting models. Optional components add a separate events service for scheduled and webhook-triggered runs, and Evolve-based memory for reusing task guidance.
Developers can use CUGA from the CLI or as a Python SDK through the CugaAgent and CugaSupervisor classes. It is built on Python 3.12 and LangGraph.
Key features
- 01Connects tools via MCP servers, OpenAPI services and Python functions
- 02Five policy types: Intent Guard, Playbook, Tool Approval, Tool Guide and Output Formatter
- 03ToolGuard checks tool calls before execution and blocks violations
- 04Supports OpenAI-compatible private inference endpoints and air-gapped deployment
- 05Multi-agent coordination through CugaSupervisor
- 06Run receipts report token usage, model calls, tool calls and elapsed time
Use cases
- Automating CRM and revenue operations with approval gates on account updates
- Answering questions over company documents with cited sources
- Coordinating specialized agents for cross-application workflows
- Running agents in a private cloud or air-gapped Kubernetes environment
Quick start
- 1
Clone the repository
git clone https://github.com/cuga-project/cuga-agent.git - 2
Enter the project directory
cd cuga-agent - 3
Install dependencies with uv on Python 3.12
uv sync --python 3.12 - 4
Start the manager with CRM tools
uv run cuga start manager --crm
Things to know
- Configuring a model provider and credentials in .env is required before the agent can run.
- Full policy, memory, and events features depend on optional services and additional configuration.
In collections
Topics
FAQ
- What is cuga-agent?
- Enterprise agent harness that turns business requests into actions across apps, APIs and documents, with policy controls and private deployment options. CUGA is an open-source generalist agent harness aimed at enterprise automation. It executes multi-step tasks across web interfaces, APIs, and documents, and connects tools through MCP servers, OpenAPI services, and Python functions. Teams can configure agents and supervisors in a browser-based management UI, test a draft, and publish a version.
- Is cuga-agent free to use? What license does it have?
- cuga-agent is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
- Is cuga-agent still maintained?
- Actively maintained. The last commit to cuga-agent was on Nov 5, 2026.
- What can I use cuga-agent for?
- Automating CRM and revenue operations with approval gates on account updates Answering questions over company documents with cited sources Coordinating specialized agents for cross-application workflows Running agents in a private cloud or air-gapped Kubernetes environment
- How do I install cuga-agent?
- Clone the repository: git clone https://github.com/cuga-project/cuga-agent.git Enter the project directory: cd cuga-agent Install dependencies with uv on Python 3.12: uv sync --python 3.12 Start the manager with CRM tools: uv run cuga start manager --crm
- Where can I get cuga-agent?
- The source code and installation instructions are on GitHub: https://github.com/cuga-project/cuga-agent
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
A Python library and runnable Jupyter textbook implementing 35 agentic AI architectures, from Reflexion and ReAct to MemGPT and BrowserAgent, on LangGraph.
A Python CLI that wraps coding agents like Codex and Claude Code with an optional supervisor, reviewer, adversary, and local log distiller, each on its own model.
A workspace where AI coding agents from different machines and tools share threads, files, and a browser in one place.
A Java library for building stateful, multi-agent LLM workflows as graphs, working with LangChain4j and Spring AI.
A Python library for building swarm-style multi-agent systems on LangGraph, where agents hand off control to each other by specialty.