Quickstart

Build your first agent in 5 minutes. Define it in YAML, test locally, and publish to the registry.

You'll need the rush CLI installed:

curl -fsSL https://cdn.getrush.ai/install.sh | bash

The same binary scaffolds, runs, builds, and publishes agents — no second CLI needed. To unlock step 5 (build and publish), enable Developer Mode once on your account.

Install the Prix Agent Builder Skill#

Let your coding assistant (Claude Code, Codex, Cursor, Gemini, OpenCode) scaffold, debug, and publish agents for you. The skill bundles the full workflow, file formats, and validator gotchas into one SKILL.md.

agents-cli

agents skills add https://prix.dev/prix-builder/SKILL.md

Vercel skills installer

npx @vercel/skills install https://prix.dev/prix-builder/SKILL.md

Manual (Claude Code or any agent with a skills directory)

mkdir -p ~/.claude/skills/prix-builder
curl -fsSL https://prix.dev/prix-builder/SKILL.md \
  -o ~/.claude/skills/prix-builder/SKILL.md

Raw file: prix.dev/prix-builder/SKILL.md. Drop it into any skills directory that reads YAML-frontmatter markdown.

What is an Agent?#

An agent is an AI-powered application that runs on Rush. Unlike chatbots that just respond to messages, agents have tools, memory, and personality — they can search the web, call APIs, generate files, remember user preferences, and delegate tasks to other agents.

Agents are defined in YAML files. No Python, no JavaScript, no Docker. Write a config, define a prompt, and your agent is ready to run.

Your First Agent#

Let's build a research agent that searches the web and delivers answers with citations.

Create the agent directory

mkdir my-researcher
cd my-researcher

Define the agent

agent.yaml

name: my-researcher
title: My Researcher
subtitle: Quick answers with sources
description: A research agent that searches the web and delivers concise answers with citations.
version: 1.0.0
developer: YourName
category: Research

models:
  anthropic/claude-sonnet-4:
    provider: openrouter

tools:
  - name: web_search
  - name: web_fetch
  - name: artifacts

Write the prompt

prompt.yaml

version: 2
role: user
sections:
  - name: identity
    content: |
      You are a research assistant. You find accurate, well-sourced
      answers to questions. You don't guess — you search, verify,
      and cite your sources.

  - name: workflow
    content: |
      1. Understand the user's question
      2. Search the web for relevant information
      3. Read and verify the most promising sources
      4. Write a clear, concise answer with inline citations
      5. Save the answer as an artifact if it's substantial

  - name: constraints
    content: |
      - Always cite sources with URLs
      - If you can't find a reliable answer, say so
      - Prefer recent sources over older ones

Test locally

rush run ./my-researcher --prompt "What is quantum computing?" --non-interactive

Add --verbose --stream-events to see tool calls and token usage in real time.

Build and publish

# Validate your agent (fmt runs automatically)
rush build ./my-researcher

# Publish to the registry
rush publish ./my-researcher

rush build auto-formats your YAML files before validation. Pass --no-fmt to skip.

Publishing requires store assets (logo, cover image) and Developer Mode. See the Publishing guide and Developer Mode.

Expose it as an API (optional)

Turn your agent into a live A2A endpoint. Local, for testing with another client:

rush serve ./my-researcher
# -> http://127.0.0.1:52341/v1 (bearer token printed)

Or cloud-hosted with a public HTTPS URL:

rush deploy your-handle/my-researcher
# -> https://agents.prix.dev/s/<id>/v1 (bearer token printed, 24h TTL)

Both speak A2A v0.3 — the same protocol LangGraph, CrewAI, Rivet, and Zed already use. See the Serve & Deploy guide for details.

What's Next?#

You've built a basic agent. Here's where to go deeper:

Agent Structure

All agent.yaml fields, prompt.yaml, ux.yaml, and assets.

Prompt Design

Write effective prompts with the mindset-over-rules approach.

Tools

All tool categories — web, storage, HTTP, bash, audio, UI.

Testing

Run locally, test subagents, and debug issues.

Publishing

Build, validate, and publish to the registry.

Serve & Deploy

Expose any agent as a live A2A endpoint — local or cloud.

Multi-Agent

Delegation, orchestrators, and system agents.