Coded agents in Studio Web (Preview)

Remarque :

Coded agents in Studio Web is in Preview.

Coded agents let you define agent behavior directly in Python while still benefiting from UiPath orchestration, evaluation, and observability. Studio Web acts as the control plane, while your Integrated Development Environment (IDE) remains the primary development environment.

After connecting your local project, Studio Web provides access to:

• Definition — A high-level view of the agent entry points and configuration.
• Source code — A read-only view of synced files. Your IDE remains the authoritative source.
• Evaluation sets — Test cases for validating agent behavior.
• Evaluators — Output- and trajectory-based evaluators used during evaluation runs.

From Studio Web you can run Debug or Debug step-by-step, execute evaluations in the Evaluations space, and publish the agent for production use. Published agents emit traces visible in the Agent traces space and can be managed from the Managing UiPath agents page.

Prérequis

• You have Python 3.11 or higher and either pip or uv installed.
• You have authenticated with UiPath Automation Cloud by running uipath auth in your project directory.
• You have created, initialized, and tested your coded agent project locally. Follow the quick start guide for your framework:
  • Core Python SDK
  • LangChain (LangGraph)
  • LlamaIndex
  • Agents OpenAI
• You have access to Studio Web with a solution open or ready to create.

Create a coded agent in Studio Web

1. Accédez à studio.uipath.com.
2. Sélectionnez Créer nouveau, puis sélectionnez Agent.
3. Select Coded as the agent type.

Result: Studio Web creates a coded agent shell and opens the Setup your coded agent page.

Connect your local project to Studio Web

On the Setup your coded agent page, Studio Web guides you through linking your local Python project to the agent.

Copy the project key

1. On the Setup your coded agent page, copy the Agent project key. Studio Web generates this key automatically for each solution project.

2. In your local project, open the .env file.

3. Add the following line, replacing <project-id-from-studio> with the key you copied:

   UIPATH_PROJECT_ID=<project-id-from-studio>
   

This key links your local project to the agent in Studio Web.

Push your project to Studio Web

1. Open a terminal in your local project directory.

2. Exécutez la commande suivante :

   uipath push
   

Result: Studio Web syncs the following files from your local project and automatically loads the coded agent content:

• Source files (for example, graph.py, entry-points.json)
• Configuration files (pyproject.toml, uipath.json)
• Evaluation assets (evaluators, evaluation sets)
• Metadata and bindings