// the find
vstorm-co/full-stack-ai-agent-template
Full-stack AI app generator — FastAPI + Next.js with AI Agents, RAG, streaming, auth, and 20+ integrations out of the box.
A cookiecutter-based CLI generator that scaffolds a FastAPI + Next.js 15 app pre-wired with your choice of AI framework, vector store, task queue, and billing. You answer a wizard, get a project with only what you chose. Aimed at teams who want to skip the boilerplate phase of building an AI product and start from something production-adjacent.
The generator model is the right call — you get code for the dependencies you actually selected, not a monolith of unused integrations sitting in your tree. PydanticAI integration uses proper typed dependency injection rather than global state, which makes testing feasible and keeps agent code readable. The multi-environment Docker split (dev with bind mounts, staging without, prod with external Nginx) is a thoughtful separation that saves the usual 'works in dev, broken in prod' Docker configuration archaeology. Shipping a real .claude/ toolkit — agent skills, slash commands, and architecture rules — is a genuinely useful addition for keeping an AI coding assistant from going off-spec on the generated project.
Five AI frameworks plus four LLM providers plus four vector stores is a combinatorial explosion of template paths. Unusual combinations (Taskiq + Qdrant + DeepAgents, say) are almost certainly undertested, and Cookiecutter templates with this many conditionals accumulate if/else cruft quickly. The pydantic-deepagents framework option is the maintainer's own library with a fraction of PydanticAI's adoption — picking it means betting on an ecosystem controlled by the same people maintaining this template, which is fine until it isn't. The '100% coverage' badge applies to the generator repo, not the generated projects — the scaffolded app's test quality depends on how much your actual business logic fits the placeholder tests. Bun, uv, and Docker are baked in deeply enough that moving off any of them after scaffolding is a non-trivial refactor.