Linear workspace with label permissions
You need a Linear workspace where you can create labels. AI-Implement relies on the
AI-Implement label to identify issues to process. You do not need admin access to the workspace — member-level access with the ability to create labels is sufficient.GitHub organization or account, and a GitHub App you control
AI-Implement uses a GitHub App to authenticate workflow dispatches and to open pull requests under the app’s identity (which is what allows the
/ai-implement comment trigger to work correctly). You need either:- A GitHub organization where you can create a GitHub App, or
- A personal GitHub account with the ability to create a GitHub App
Somewhere to run the orchestrator
The orchestrator is a small Node.js service. You have two options:
Fly.io (recommended)
A Fly.io account lets you deploy the orchestrator as a managed app. It runs comfortably on a
shared-cpu-1x machine with 256 MB of memory. See Deploy the orchestrator for instructions.Any server running Node.js
You can run the orchestrator on any server or local machine that has Node.js 20 or later installed. The local development workflow is covered in Deploy the orchestrator.
A Claude provider
AI-Implement supports three ways to authenticate with Claude. You need at least one:
| Provider | What you need |
|---|---|
| Anthropic API (fallback) | An ANTHROPIC_API_KEY from console.anthropic.com |
| Claude Code OAuth (preferred) | A CLAUDE_CODE_OAUTH_TOKEN — preferred over the API key when both are present |
| AWS Bedrock | An AWS account with Bedrock access and an IAM role configured for GitHub OIDC |
You configure the provider per target repo, not globally. If you are just getting started, an Anthropic API key is the simplest path.
What’s next
Once you have everything above, work through the setup in order:Create a GitHub App
Create and configure the GitHub App that AI-Implement uses to authenticate with GitHub and open PRs.
Deploy the orchestrator
Run the orchestrator locally or deploy it to Fly.io.