Build coded agents with Coding Agents

A coding agent with the uipath-agents skill installed can scaffold, write, package, and publish Python-based agents to UiPath Orchestrator entirely from your terminal. Coded agents are standard Python projects built with LangGraph, LlamaIndex, or OpenAI Agents, and UiPath handles orchestration, scheduling, escalation, and governance once the agent is deployed.

For setup (installing the agent, logging in, installing skills), see Install and set up.

For a complete reference of the uip codedagent commands, see the UiPath CLI guide.

Voraussetzungen

• The uipath-agents skill installed for your coding agent. See Install and set up.
• Eine angemeldete Sitzung (uip login).
• Python 3.11 or newer.
• The uv package manager.

Was der Agent tun kann

Area	Was der Agent tun kann	Beispiel:
Initialize a project	Create a new Python project with uv, add the UiPath SDK package of your choice, and run uip codedagent setup and uip codedagent new to scaffold the project structure	"Initialize a new coded agent project called invoice-processor."
Write agent logic	Write and iterate on Python agent code using LangGraph, LlamaIndex, or OpenAI Agents	"Add a tool that calls the UiPath context grounding index to check company policy."
Konfigurieren	Run uip codedagent init to wire the Python project to UiPath's orchestration layer	"Configure the agent for UiPath Orchestrator."
Paket	Run uip codedagent pack to bundle the agent as a deployable artifact	"Package the agent."
Veröffentlichen	Run uip codedagent publish to upload the package to Orchestrator and auto-create a process	"Publish the agent to my personal workspace."
Sync with Studio Web	Run uip codedagent push to upload local files to the remote Studio Web project, or uip codedagent pull to download all files from the remote project to your local workspace	"Push my local changes to Studio Web."

The build loop

The coding agent follows this cycle when building or iterating on a coded agent:

Initialize > Write > Evaluate > Package > Sync with Studio Web (optional) > Publish > Set environment variables > Run

1. Initialize — create a new Python project, add the UiPath SDK, and scaffold the project structure:

   mkdir my-agent && cd my-agent
   uv init
   uv add uipath-langchain   # or uipath-llamaindex / uipath-openai-agents
   uip codedagent setup
   uip codedagent new my-agent
   uip login
   uip codedagent init
   

2. Write — ask the coding agent to write and edit the Python agent logic. Coded agents are standard Python and are not limited to UiPath-specific capabilities.

3. Evaluate — test the agent before publishing:

   • Analyze — ask the coding agent to set up a test set, run it, and explain the scores.

   • Iterate — refine the prompt or logic until the results look right.

4. Package — package the agent with uip codedagent pack to create a deployable artifact.

5. Sync with Studio Web (optional) — if you want to make further changes in Studio Web:

   • Push — run uip codedagent push to upload local files to the remote Studio Web project.
   • Pull — run uip codedagent pull to download all files from the remote Studio Web project to your local workspace.

6. Publish — publish the agent with uip codedagent publish --my-workspace to upload the package to Orchestrator. A process is auto-created in your personal workspace on first publish.

7. Set environment variables — in Orchestrator, open the process configuration and add any required environment variables (API keys, connection strings).

8. Run — start the job.

Result: The agent is deployed to Orchestrator and running as a process. UiPath handles scheduling, retries, and logging.

SDK packages

Choose the package that matches your agent framework:

Paket	Framework
uipath-langchain	LangChain / LangGraph
uipath-llamaindex	LLamaIndex
uipath-openai-agents	OpenAI Agents

All three packages expose the same UiPath capabilities (context grounding, human-in-the-loop escalation, guardrails, Job attachments) through framework-native abstractions.

Example builds

Open your coding agent and describe the agent you want to build. Because coded agents are standard Python, there are no limits on what you can ask for — the UiPath SDK adds orchestration, governance, and platform integration on top.

What to ask	What gets built
"Create a coded agent that researches a topic and produces a summary report."	Research agent with LLM calls
"Build a coded agent that analyses a CSV file given as input."	Agent wired to receive a Job attachment as input
"Create a coded agent that uses a UiPath context grounding index to check company policy."	RAG (retrieval-augmented generation) pipeline against a UiPath context grounding index

Beispiele

The following repositories contain reference implementations:

• UiPath LangChain samples — 20+ samples covering guardrails, human-in-the-loop, distributed agents, and conversational agents.
• LlamaIndex samples — UiPath integration patterns for LlamaIndex.
• OpenAI Agents samples — UiPath integration patterns for OpenAI Agents.

Siehe auch

• Install and set up — agent setup and skills install.
• Build low-code agents with Coding Agents — build Agent Builder agents from a coding agent.
• UiPath Python SDK documentation — full SDK reference for coded agents.