Bring your framework
Write agents in LangChain, LlamaIndex, or the OpenAI Agents SDK. They run on UiPath unchanged, with no rewrite and no proprietary framework.
Build agents in code, in your IDE
Write agents in Python with the UiPath SDK and CLI, on frameworks like LangChain and LlamaIndex, or accelerate with UiPath for Coding Agents. UiPath runs the orchestration, governance, and observability underneath.
Build Agents with Code
Build agents in code, in your IDE
Write agents in Python with the UiPath SDK and CLI, on frameworks like LangChain and LlamaIndex, or accelerate with UiPath for Coding Agents. UiPath runs the orchestration, governance, and observability underneath.
An enterprise AI agent framework that runs the Python you already write
Building a Python AI agent is the easy part. Getting it to run reliably in governed production is where most projects stall.
Build in Python, in your IDE
Develop in Python in your own IDE, on open-source foundations. The UiPath SDK connects your code to the platform.
Accelerate from a prompt
UiPath for Coding Agents lets developers direct a coding agent to build, troubleshoot, test, and deploy agents on UiPath without leaving the IDE.
Deploy to governed production
Deploy through the same pipeline and runtime that already runs mission-critical automation across the enterprise.
A code-first development experience, end to end
Write in Python, test against ground truth, and ship through your own pipeline. The UiPath platform handles how the agent runs in production.
Agents reach the systems where the work happens
Agents call RPA robots, IXP, and enterprise integrations as tools, reaching SAP, Citrix, and legacy apps with no API alongside document-heavy unstructured workflows. The systems where the real work happens, not just the ones with an API.
Write agents in Python
Develop in Python with the UiPath SDK and CLI. Build agents directly against platform services from your own code.
Bring agents built elsewhere
Already built agents on another framework or platform? Run them on UiPath under the same governance and orchestration as everything else.
Direct a coding agent from your IDE
UiPath for Coding Agents gives coding agents the UiPath artifact knowledge and context to build, troubleshoot, test, and deploy agents from a prompt without leaving your development environment.
Test and optimize before you ship
Run evals against ground truth before deploy. Score quality, catch drift between deploys, and let Optimize propose targeted changes to the agent.
Ship through one pipeline
Pack agents as part of your normal CI/CD. Deploy through Orchestrator alongside the robots and automations your team already manages.
Run governed in production
Maestro orchestrates, the AI Trust Layer governs, traces and audit logs make it observable. Independently certified to ISO 42001 and AIUC-1.
See developers build and run AI agents in code
Watch the code-first workflow end to end, and how agents reach production with MCP, A2A, and the platform underneath.
What running on UiPath changes for your agents
The same agent you build in Python runs in production. Here is what you get that an open-source framework leaves you to build yourself.
Full control
Choose your model, framework, and reasoning approach. UiPath does not dictate how the agent thinks. It makes sure it runs reliably in production.
Stay in your workflow
Existing repos, CI practices, review processes, and coding conventions stay intact. Your team keeps building the way it already works.
No platform plumbing
Orchestration, governance, observability, and deterministic execution are handled by the platform. Spend your time on agent logic instead.
Production-ready by default
UiPath ships retries, recovery patterns, observability, and evals as part of the runtime, not as code your team writes and maintains. Ready on day one, not after a hardening project.
Start building in code
Install the UiPath SDK and CLI to build agents in Python on open-source foundations, with observability, version control, and one-command deployment built in.
UiPath Python SDK
Install the SDK and CLI. Programmatic access to assets, buckets, Context Grounding, and the LLM Gateway.
LangChain
This guide provides step-by-step instructions for setting up, creating, publishing, and running your first UiPath-LangChain Agent.
LlamaIndex
This guide provides step-by-step instructions for setting up, creating, publishing, and running your first UiPath-LlamaIndex Agent.
OpenAI Agents
This guide provides step-by-step instructions for setting up, creating, publishing, and running your first UiPath OpenAI Agent.
Built with the ecosystem developers trust
We've seen a key part of this be AI observability, and we're excited to integrate LangSmith with UiPath to help even more builders ship agents with confidence.
WHITE PAPER
Success in the Age of Composable AI Agents
Where agentic AI is heading, and what it takes to run composable agents reliably in enterprise production.
More ways to build and run agents
Pick the build experience that fits the project. Every path runs on the same governed platform.
Build with low-code
Drag-and-drop canvas in Agent Builder for the people closest to the process.
Build with UiPath for coding agents
Direct AI coding agents like Claude Code, Codex, or Cursor from your IDE or CLI.
Connect any AI model
Use any LLM, first-party or third-party, with model usage policies, data residency controls, and PII redaction applied uniformly via the AI Trust Layer.
Get it all SaaS
With Automation Cloud™, you get the full UiPath Platform™, the fastest updates—and the scalability and flexibility to automate any process, anywhere you do business.
Start here
Need to self-host?
Run the platform yourself, with the world-class automation capabilities you expect and the performance, compliance, and security you require.
Start hereBuilding agents in code: common questions
Answers for developers evaluating where to build and run production AI agents.
No. LangChain and LlamaIndex are the frameworks. UiPath is the enterprise runtime that takes those agents to production unchanged, adding observability, version control, governance, and Maestro deployment.
Model-agnostic by design. Connect Anthropic, OpenAI, Google, open-source, or UiPath-hosted models. The same agent code switches model backends without a rebuild, and the AI Trust Layer applies regardless of which model it talks to.
Agents run inside the platform that already orchestrates RPA, so Python agents call robots, document processing, and integrations as tools. SAP, Citrix, and desktop apps included, not just API-based systems.
Yes. Agents work as Model Context Protocol clients and hosts, so they can use any MCP-compatible tool, and the UiPath platform surface is exposed over MCP for other agents to consume.
They are peer choices, not a hierarchy. Build low-code in Studio when the people closest to the process are building. Build in code when developers want deeper customization. A coding agent can accelerate either path via UiPath for Coding Agents. An agent built low-code can move to code at any point without a rebuild.
Evals score outcomes against ground truth before and after deploy, traces and audit logs make every step inspectable, and Action Center routes exceptions to a human rather than letting them fail silently.