CUGA
for real enterprise work
CUGA is the open-source agent runtime behind the Crafted AI Framework. Start with a benchmark-proven generalist harness, then configure domain tools, MCP servers, policies, reasoning modes, and deployment paths without rebuilding planning, orchestration, browser control, and governance from scratch.
Core Capabilities
The pieces enterprises usually have to assemble themselves, already wired into a configurable harness.
OpenAPI, MCP & LangChain tools
Wire existing REST APIs, MCP servers, and Python functions into one agent runtime without rebuilding orchestration.
Reasoning modes
Tune CUGA for fast, balanced, accurate, or experimental save-and-reuse runs depending on cost and reliability needs.
Hybrid API + browser execution
Use APIs for system data, then operate web interfaces with Playwright and the CUGA browser extension in the same task.
Policies and human approval
Configure Intent Guard, Playbook, Tool Approval, Tool Guide, and Output Formatter policies with HITL checkpoints.
Supervisor and remote agents
Compose CUGA as a worker, supervisor, or remote A2A collaborator for nested workflows and multi-agent execution.
Manage, publish, and audit
Draft tools, MCP servers, LLM choices, and policies in the manager UI, test them, then publish versioned configs.
CUGA is the engine layer. Crafted turns it into shipped agents.
The Crafted AI Framework does not start from a blank agent scaffold. It uses CUGA as the governed runtime underneath, then adds the product work around it: domain modeling, tool contracts, deployment topology, evaluation loops, and the user-facing control plane.
The execution engine
CUGA supplies the agent loop: planning, tool selection, API calls, browser actions, memory hooks, approvals, and run visibility.
The product layer
Crafted wraps that runtime with reusable deployment patterns, domain templates, evaluations, operating surfaces, and managed implementation support.
The business system
Your tools, policies, data, and user experience become a governed agent that can run inside real enterprise processes.
Enterprise Use Cases
CUGA is strongest where the task crosses systems, interfaces, policies, and steps.
API-to-web workflows
Fetch account or operational data from APIs, then update the right web page, form, or internal console.
Tool-rich enterprise agents
Coordinate OpenAPI services, MCP tools, and runtime LangChain functions in a single task plan.
Governed execution
Pause at critical decisions, inspect planned actions, and require approval before sensitive tools run.
Reusable domain playbooks
Package domain procedures as SKILL.md files so the agent loads full instructions only when a task needs them.
Planner-executor + CodeAct
Structured planning, code execution, and smart variable management help CUGA handle long, multi-step tasks.
SDK and CLI friendly
Run demos with the CLI, embed CUGA from Python, or pass runtime tools directly when CUGA lives inside your app.
Optional memory
Enable memory when the workflow needs continuity, then keep lighter stateless runs for simpler automation.
Langflow integration
Use CUGA in low-code visual workflows when teams need to design and ship agent flows quickly.
Cluster-ready self-hosting
Deploy with included Docker assets, Helm chart, and scripts for local or cloud Kubernetes clusters.
Trajectory visibility
Inspect agent trajectories, tool calls, and decisions with the visualization dashboard after a run.
Why CUGA?
A practical comparison for teams deciding whether to build an agent stack or start with a harness.
Put CUGA behind your Crafted AI Framework deployment.
Crafted helps teams turn the open-source CUGA harness into production domain agents: tool wiring, policy design, evaluation, deployment, and the operating surface your users actually need.
Plan a CUGA Deployment