Arkitekt.ai

A project generator for AI agents, in the spirit of Spring Initializr. Pick a language, an agent framework, a model provider and a few capabilities, then download a project that installs and runs.

What you can generate

Option	Choices
Language	Python, TypeScript
Framework	Provider SDK (both), LangGraph (Python), Vercel AI SDK (TypeScript)
Provider	Anthropic, OpenAI, Ollama
Capabilities	Tools, conversation memory, RAG, MCP client, HTTP API, Docker, tests

The page also shows every generated file before you download, and the setup lives in the URL so you can share it.

API

# Options and their allowed values live in src/lib/catalog.ts
curl -X POST http://localhost:3000/api/starter.zip \
  -H 'content-type: application/json' \
  -d '{"name":"my-agent","language":"python","framework":"langgraph","provider":"anthropic","model":"claude-sonnet-5-5","capabilities":["tools","memory"]}' \
  -o my-agent.zip

PUT /api/starter.zip takes the same body and returns the files as JSON. GET /api/metadata lists the options.

Versioning

The version lives in package.json and shows in the page footer, in GET /api/metadata and in every generated README. See CHANGELOG.md.

Develop

npm install
npm run dev

Templates are code, in src/lib/generators/. There is no database: every request builds the zip from scratch, so the app deploys to Vercel as is.

Check the generated projects

npm run verify -- python langgraph anthropic all
npm run verify -- typescript vercel-ai openai min

This generates the project, installs it, type-checks or compiles it, imports the agent and runs its tests. CI runs 16 combinations on every push.