Create your first agent.

Install Skybert, connect your LLM provider through Pi, then use the Skybert CLI in your terminal to build your agent.

1

Install the Python SDK

Add Skybert to the environment where your agent runs.

uv add skybert

Recommended for modern Python projects.

2

Install Pi and connect your model provider

Skybert uses Pi for the guided build flow. Install Pi, then use Pi's provider login for the model stack your agent will use.

terminal
npm install -g --ignore-scripts @earendil-works/pi-coding-agent
pi
/login
Model provider setup: Pi provider docs.
3

Launch the Agent Builder

Create a new folder for your agent and then simply launch skybert in that folder to start building. If you have any documents or data that you think might be useful for Skybert to understand your business need, just drop these files in that folder before launching.

terminal
mkdir my-first-agent
cd my-first-agent
uv init
uv add skybert
skybert
4

Choose Build new agent

The fullscreen workspace opens locally. Choose Build new agent, answer the prompts, and let Pi create the agent code with Skybert instrumentation.

If you already have a working agent that you would like skybert to Optimize, select Add existing agent instead. Make sure you launch Skybert from the same folder that contains your agent code.

Skybert local workspace with Build new agent selected
Normal runs stay local and quiet. Skybert captures observations in .skybert/; improvements happen as code changes that you ship through git.
+

Connect to Skybert Cloud

Optional

Cloud is not required to build locally. Connect when you want to collect production data, annotate datasets, team collaboration, and version monitoring.

skybert login
Open Cloud Dashboard