Skip to main content

Quick Start

Get up and running with GenieBuilder in minutes.

1. Complete Onboarding​

When you first launch GenieBuilder, the setup wizard will guide you through:

  • Selecting a project workspace
  • Configuring cloud AI providers (OpenAI, Anthropic, Gemini)
  • Setting up local AI (Ollama) if desired
  • Configuring MCP tools

You can skip any step and configure it later in Settings.

2. Create Your First Backlog Task​

GenieBuilder uses Backlog.md for task management. Workflows run in the context of a task, so creating one is the first step before automating any work.

To create a task:

  1. Open the chat panel and ask: "Create a task: Add a greeting message to the homepage"
  2. The Backlog MCP tool will create the task and assign it an ID (e.g., TASK-1)
  3. Your new task appears in the Task Board panel on the right

You can also create tasks directly from the Task Board by clicking + New Task.

tip

Workflows use task details (title, description, acceptance criteria) as context for each step. A well-written task leads to better automated results.

3. Open a Workspace​

Select a project folder using the file explorer. GenieBuilder will:

  • Show a .gitignore-aware file tree
  • Allow you to open files in the built-in editor
  • Track tasks in the right panel (if Backlog is configured)

4. Configure an AI Provider​

Choose from cloud or local providers:

Cloud Providers:

  • OpenAI — GPT-5.4 and other GPT models
  • Anthropic — Claude 4.6 Opus or Sonnet
  • Google — Gemini 3.1 Pro and Flash

Local Providers:

  • Ollama — Run models locally for privacy
  • LM Studio — GUI-based local model management
  • llama.cpp — Direct integration with custom builds

Add your API keys in Settings → Providers. Keys are stored securely using macOS Keychain.

5. Start a Chat​

  1. Open the chat panel (bottom of screen)
  2. Click New Tab to start a conversation
  3. Select your provider and model
  4. Type your prompt and press ⌘+Enter to send

Each chat tab can use a different provider, letting you compare responses or use specialized models for different tasks.

6. Try a Workflow​

Workflows let you chain multiple AI agents into an automated pipeline. A great starting point is the "Implement, Review, Fix Issues" example — a multi-agent workflow that uses different CLI Runners for each stage:

StepCLI RunnerRole
ImplementClaudeWrites the code based on the task description
ReviewCodexReviews the implementation for bugs and improvements
Fix IssuesClaudeAddresses any issues found during review
Git OperationsKimiCommits changes, pushes the branch, and opens a PR

This pattern works well because each agent brings different strengths — Claude for deep implementation, Codex for thorough review, and Kimi for fast git operations.

To try it:

  1. Click the Workflow button in the header
  2. Select "Implement, Review, Fix Issues" from the examples sidebar
  3. Pick a task from the Task Board when prompted
  4. Click Run and watch the agents work through each stage

Implement, Review, Fix Issues workflow

Click any node to configure its executor type, CLI Runner, tool access, and prompt template:

Workflow node settings

After a workflow completes, open the Runs panel to inspect the full execution history. Each run shows the status of every step, timing, the CLI Runner used, and the complete output — so you can see exactly what each agent did:

Workflow runs monitor

Next Steps​

Last updated on