v0.1 · MIT License

Portable AI
Tooling Setup

One canonical YAML. Apply to Codex, Claude, OpenCode, or Gemini — without rewriting a single rule.

Get Started GitHub
terminal
$ npm install -g @agentstack/cli + @agentstack/cli@0.1.0 $ agentstack init ✓ Canonical profile created → agentstack.yaml

Everything you need to move fast

Stop configuring AI tools from scratch every time you switch.

Portable

One YAML, any AI tool. Switch providers without rewriting rules or instructions.

Composable

Build, share and install profiles from GitHub. Community-driven tooling templates.

Secrets-safe

API keys never in plaintext. Backed by OS keychain or encrypted file backend.

Multi-provider

Native support for Codex, Claude, OpenCode and Gemini — with more on the way.

Sync-enabled

Zero-config sync keeps your setup consistent across machines and teammates.

Open

MIT-licensed CLI. No SaaS dependency, no lock-in. Your config, your way.

From existing setup to portable profile in minutes

01

Import

agentstack import --from codex --source ~/.codex/

Capture your existing AI tool configuration into a canonical profile.

02

Customize

# edit agentstack.yaml

Add instructions, MCP servers, and skills to your canonical profile.

03

Apply

agentstack apply --to claude

Deploy your profile to any supported AI tool — instantly.

agentstack.yaml 
version: 3
profile:
  name: my-dev-setup
  description: My portable AI coding environment

  providers:
    claude:
      enabled: true
    codex:
      enabled: true
    opencode:
      enabled: false

  instructions:
    global: |
      You are a senior TypeScript developer.
      Prefer functional patterns and explicit types.
      Always write tests for new code.

  mcp_servers:
    - id: filesystem
      transport: stdio
      command: npx
      args: ["-y", "@modelcontextprotocol/server-filesystem", "."]
      enabled: true

  permissions:
    command_policy: on-request

One file, infinite portability

Your AI tooling setup, canonicalized

  • → Providers — enable or disable per tool
  • → Instructions — global rules applied everywhere
  • → MCP servers — portable tool definitions
  • → Secrets-safe — env_refs keep keys out of YAML
View full schema →

Works with your tools

Supports all major AI coding tools

C
Codex OpenAI
supported
A
Claude Anthropic
supported
O
OpenCode Community
supported
G
Gemini Google
supported

One canonical profile — applied to any tool with a single command.