#657 · Primary category: AI Agents & Automation

cuga-agent

coding-agent computer-use enterprise generalist-agent guardrails harness knowledge mcp policies sandbox

CUGA is an open-source generalist agent harness for the enterprise, supporting complex task execution on web and APIs, OpenAPI/MCP integrations, composable architecture, reasoning modes, and policy-aware features.

Project last updated:08/28/26

GitHub Stars

875

Forks

147

Contributors

26

License

Apache-2.0

Why we included this project

Most of the work in building an enterprise agent in-house is scaffolding: wiring tools, layering in planning logic, adding guardrails, and then measuring cost and reliability. CUGA packs that into a configurable harness, so you can point a generalist agent at your OpenAPI specs, MCP servers, or LangChain tools and tune it per domain. A single flow can mix API calls with browser work through Playwright, and supervisor mode hands larger jobs off to subtasks, optionally to remote agents. For regulated settings, safety is configuration rather than a code fix; intent guards, playbooks, tool-approval gates, and human-in-the-loop checkpoints translate policy into enforced runtime behavior. It comes as a Python SDK, a self-hostable service with a Helm chart, an MCP server, and a Langflow component, so a small proof of concept can grow into a deployed cluster without rebuilding orchestration.

Articles for this project

No articles for this project yet.

To suggest a topic or contribute an article, contact us.

Related projects in this category