- Getting started
- UiPath Agents in Studio Web
- About UiPath Agents
- Licensing
- Running agents
- Working with files
- Contexts
- Escalations
- Evaluations
- Agent traces
- Agent score
- Managing UiPath agents
- UiPath Coded agents
- Build with Coding Agents
Build UiPath agents entirely from a coding agent — no browser required after the initial login. Describe the agent you want, and the installed UiPath skills guide your coding agent through the full lifecycle: scaffolding, configuring, testing, and deploying.
How it works
Skills deliver curated UiPath knowledge
UiPath ships dedicated skills for agent development. A skill is a package of instructions and context that teaches your coding agent exactly how to do UiPath work — the right file structure, the right CLI commands, the validation loops, and the platform conventions. Without the skill, the coding agent guesses; with it, the coding agent knows.
One skill covers all agent development:
uipath-agents — the full lifecycle for both agent types. The skill automatically picks the correct path based on your prompt:
- Low-code agents: scaffold, configure prompts and schemas, add tools, validate, upload to Studio Web, evaluate, and deploy.
- Coded agents: initialize a project, write agent logic with LangGraph, LlamaIndex, or OpenAI Agents, package, and publish to Orchestrator.
Works with any major coding agent
UiPath skills work across the main AI coding agents:
| Coding agent | Form factor |
|---|---|
| Claude Code | Terminal and IDE |
| Cursor | IDE |
| Codex CLI | Terminal |
| Google Antigravity | Agentic development platform |
The UiPath experience is consistent across all of them — the same skills, CLI commands, and project conventions apply regardless of which coding agent you use.
Set it up in one command
Install the uip CLI, the skills, and the required runtimes with a single one-liner. For the full setup walkthrough, see Install and set up.
For a complete reference of the uip agent commands, see the UiPath CLI guide.
What you can build
The following eight steps describe the full developer journey for building a UiPath agent with a coding agent. Steps 1 and 2 determine the agent type; the remaining steps apply to both.
-
Choose and scaffold — Decide between a low-code agent (configure through prompts and prebuilt UiPath tools, no Python) or a coded agent (full Python control with LangGraph, LlamaIndex, OpenAI Agents, or a plain Python function). The coding agent scaffolds the project structure from a single prompt.
-
Build the agent's behavior — For low-code: define the system prompt, user message template, input/output schemas, and LLM model settings in
agent.json. For coded: write and iterate on the Python agent logic. -
Give the agent tools and capabilities — Connect the agent to what it needs: call RPA processes, use other agents as tools, add Integration Service connectors (Slack, Jira, SharePoint, and others), ground the agent on company knowledge with context grounding, accept file attachments, and make the agent conversational.
-
Add governance and human oversight — Apply guardrails (PII masking, harmful content filtering, custom rules) and define escalation points so a human can review, approve, or correct the agent at key steps.
-
Run and iterate locally — Execute the agent on sample input or open a local dev console to interact with it in the browser before anything is published.
-
Evaluate quality — Build test and evaluation sets, run them against the agent runtime, and review scores and execution traces. Catch regressions as you refine prompts or logic.
-
Deploy and publish — Package the agent and publish it to UiPath Orchestrator so it can run for real, in a personal workspace, a tenant feed, or a specific folder.
-
Integrate into the wider platform — Embed the agent as a step in a Maestro flow, use it as a tool for another agent, or bundle it into a multi-project solution so it becomes part of an end-to-end automation.
For step-by-step guidance on each agent type, see Build low-code agents with Coding Agents and Build coded agents with Coding Agents.
For a complete walkthrough from scaffold to deployment, see Example: social sentiment agent.