Full-Stack AI Agent Template

Production-ready FastAPI + Next.js project generator with AI agents, RAG, and 20+ enterprise integrations.

Quick Start • Features • Demo • Documentation • Configurator • PyPI

🤖 5 AI Agent Frameworks (PydanticAI, PydanticDeep, LangChain, LangGraph, DeepAgents)
📄 RAG Pipeline (Milvus, Qdrant, pgvector, ChromaDB)
⚡ FastAPI + Next.js 15 (WebSocket streaming, real-time chat UI)
🔗 Conversation Sharing (direct sharing, public links, admin browser)
🔒 Enterprise-Ready (JWT, OAuth, admin panel, Celery, Docker, K8s)

Vstorm OSS Ecosystem

This template is part of a broader open-source ecosystem for production AI agents:

Project	Description
pydantic-deepagents	The modular agent runtime for Python. Claude Code-style CLI with Docker sandbox, browser automation, multi-agent teams, and /improve.
pydantic-ai-shields	Drop-in guardrails for Pydantic AI agents. 5 infra + 5 content shields.
pydantic-ai-subagents	Declarative multi-agent orchestration with token tracking.
summarization-pydantic-ai	Smart context compression for long-running agents.
pydantic-ai-backend	Sandboxed execution for AI agents. Docker + Daytona.

Want the runtime behind this template's AI agents? pydantic-deepagents powers the deepagents framework option — install it standalone with curl -fsSL .../install.sh | bash.

Browse all projects at oss.vstorm.co

🚀 Quick Start

[!TIP] Prefer a visual configurator? Use the Web Configurator to configure your project in the browser and download a ZIP — no CLI installation needed.

Installation

# pip
pip install fastapi-fullstack

# uv (recommended)
uv tool install fastapi-fullstack

# pipx
pipx install fastapi-fullstack

From zero to a running app

Three steps. The wizard scaffolds the project, make bootstrap brings up the whole backend, and the frontend runs with a single command:

# 1. Generate your project — just answer the wizard's prompts
fastapi-fullstack

# 2. Backend + PostgreSQL up, migrations applied, default admin seeded
cd my_ai_app
make bootstrap

# 3. Frontend (in a second terminal)
cd frontend && bun install && bun dev

What make bootstrap does (= make dev + make seed): builds the backend Docker image, starts the stack via docker-compose.dev.yml, waits for PostgreSQL (pg_isready), applies Alembic migrations, and seeds admin@example.com / admin123. It's idempotent — re-run it anytime.

Then access:

Day-to-day commands

make dev           # bootstrap or restart (no admin re-seed)
make seed          # one-shot admin creation (no-op if admin exists)
make dev-down      # stop everything
make dev-logs      # tail container logs
make dev-rebuild   # force-rebuild backend image (after pyproject.toml changes)
make dev-frontend  # start the Next.js container

After the first make bootstrap, day-to-day you just run make dev (skips admin re-seed). Run make help inside the project for the full list.

# Non-interactive with explicit options
fastapi-fullstack create my_ai_app --database postgresql --frontend nextjs

# Presets for common scenarios (run `fastapi-fullstack templates` for the full list)
fastapi-fullstack create my_ai_app --preset ai-agent           # AI agent with streaming
fastapi-fullstack create my_ai_app --preset production         # Full production setup
fastapi-fullstack create my_ai_app --preset production-saas    # SaaS: billing, teams, admin

# Bare-bones project (PostgreSQL, no Docker/Redis/CI)
fastapi-fullstack create my_ai_app --minimal

Environments

Each env has matching -down, -logs, -rebuild siblings.

[!NOTE] Windows users: make requires GNU Make. Install via Chocolatey (choco install make) or use WSL2 / Git Bash. The Docker workflow is identical across macOS, Linux, and WSL2.

cd my_ai_app
make install                                                # uv sync + pre-commit hooks

# Start only infrastructure containers
docker compose -f docker-compose.dev.yml up -d db redis    # add 'milvus etcd minio' if RAG

make db-upgrade                                             # apply migrations
make create-admin                                           # interactive
make run                                                    # uvicorn --reload

# On your server
git clone <your-repo>
cd my_ai_app

cp backend/.env.example backend/.env            # fill in real secrets
# Configure your nginx host using nginx/nginx.conf as reference

make prod                                       # builds + starts + migrates
make prod-logs                                  # tail logs

cd frontend && npx vercel --prod

Using the Project CLI

Each generated project has a CLI named after your project_slug. For example, if you created my_ai_app:

cd backend

# The CLI command is: uv run <project_slug> <command>
uv run my_ai_app server run --reload     # Start dev server
uv run my_ai_app db migrate -m "message" # Create migration
uv run my_ai_app db upgrade              # Apply migrations
uv run my_ai_app user create-admin       # Create admin user

Use make help to see all available Makefile shortcuts.

🎬 Demo

CLI generator — configure and scaffold a full-stack AI project in under 60 seconds:

Generated marketing site — public landing page with hero, pricing, blog, and legal pages (enable_marketing_site):

📸 Screenshots

AI Chat

The chat UI streams responses over WebSocket and renders each tool call as a purpose-built card.

Auth & Dashboard

Teams & Knowledge Bases

Billing & Usage

Profile & Settings

Admin Panel

Marketing Site

Background Tasks, Observability & Channels

🎯 Why This Template

Building AI/LLM applications requires more than just an API wrapper. You need:

• Type-safe AI agents with tool/function calling
• Real-time streaming responses via WebSocket
• Conversation persistence and history management
• Production infrastructure - auth, rate limiting, observability
• Enterprise integrations - background tasks, webhooks, admin panels

This template gives you all of that out of the box, with 20+ configurable integrations so you can focus on building your AI product, not boilerplate.

Perfect For

• 🤖 AI Chatbots & Assistants - PydanticAI or LangChain agents with streaming responses
• 📊 ML Applications - Background task processing with Celery/Taskiq
• 🏢 Enterprise SaaS - Full auth, admin panel, webhooks, and more
• 🚀 Startups - Ship fast with production-ready infrastructure

AI-Agent Friendly

Generated projects include CLAUDE.md and AGENTS.md files optimized for AI coding assistants (Claude Code, Codex, Copilot, Cursor, Zed). Following progressive disclosure best practices - concise project overview with pointers to detailed docs when needed.

They also ship a ready-to-use .claude/ toolkit that adapts to the options you selected:

• Agent Skills (.claude/skills/) — model-invoked playbooks that auto-trigger when relevant: alembic-migration, pytest-suite, agent-tool (framework-aware), frontend-feature, rag-knowledge, background-task (queue-aware), billing-stripe, and channel-bot. Feature-gated — only the skills that match your stack are generated.
• Slash commands (.claude/commands/) — /add-endpoint, /fix-issue, /review.
• Convention rules (.claude/rules/) — architecture, code style, schemas, exceptions/security, testing, and frontend conventions, loaded automatically.

✨ Features

🤖 AI/LLM First

• 5 AI Frameworks - PydanticAI, PydanticDeep, LangChain, LangGraph, DeepAgents
• 4 LLM Providers - OpenAI, Anthropic, Google Gemini, OpenRouter
• RAG - Document ingestion, vector search, reranking (Milvus, Qdrant, ChromaDB, pgvector)
• WebSocket Streaming - Real-time responses with full event access
• Rich Chat UI - Specialized tool-call cards (web search, knowledge base, Python, charts, skills), live subagent feed, citation sources panel, plan/task checklist, reasoning view, and in-chat file previews
• Agent Tools - Web search, URL fetch, charts, code execution (run_python), skills, ask_user, plus optional Deep Research (TODO planner + parallel subagents)
• Messaging Channels - Telegram and Slack multi-bot integration with polling, webhooks, per-thread sessions, group concurrency control
• Conversation Sharing - Share conversations with users or via public links, admin conversation browser
• Conversation Persistence - Save chat history to database
• Message Ratings - Like/dislike responses with feedback, admin analytics
• Image Description - Extract images from documents, describe via LLM vision
• Multimodal Embeddings - Provider-aware: OpenAI, Voyage (Anthropic), Gemini (multimodal text + images)
• Document Sources - Local files, API upload, Google Drive, S3/MinIO
• Sync Sources - Per-organization connector management UI (Google Drive, S3/MinIO) with scheduled sync, manual triggers, encrypted credentials, and per-run logs
• Observability - Logfire for PydanticAI, LangSmith for LangChain/LangGraph/DeepAgents

⚡ Backend (FastAPI)

• FastAPI + Pydantic v2 - High-performance async API
• PostgreSQL (async) - SQLAlchemy 2.0 + Alembic migrations, pgvector-ready
• Authentication - JWT + Refresh tokens, API Keys, OAuth2 (Google)
• Background Tasks - Celery, Taskiq, ARQ, or Prefect
• Django-style CLI - Custom management commands with auto-discovery

🎨 Frontend (Next.js 15)

• React 19 + TypeScript + Tailwind CSS v4
• AI Chat Interface - WebSocket streaming, tool call visualization
• Authentication - HTTP-only cookies, auto-refresh, password reset, magic link
• Marketing Site - hero, pricing, FAQ, blog, contact form, legal pages (PL + EN)
• Billing Dashboard - subscription, payment methods, invoices, credits balance/ledger, and usage charts (Stripe)
• User Settings - profile, API keys CRUD (sk_* tokens), onboarding tracking
• Admin Panel - workspace stats, message-rating analytics, Stripe events browser
• SEO - per-page metadata, OG image, sitemap, robots, manifest, favicons
• Dark Mode + i18n (PL/EN via next-intl, locale-prefixed routes)

🔌 20+ Enterprise Integrations

🗺️ Architecture Overview

┌──────────────────────────────────────────────────────────────────────────┐
│                         FRONTEND  (Next.js 15)                           │
│  Chat UI · Knowledge Base · Dashboard · Settings · Dark Mode · i18n      │
└──────────────┬───────────────────────────────────────────┬───────────────┘
               │  REST / WebSocket                         │  Vercel
               ▼                                           ▼
┌──────────────────────────────────────────────────────────────────────────┐
│                         BACKEND  (FastAPI)                               │
│                                                                          │
│  ┌─────────────────────────────────────────────────────────────────┐     │
│  │                     AI AGENTS                                   │     │
│  │  PydanticAI · LangChain · LangGraph · DeepAgents                │     │
│  │  ────────────────────────────────────────────────────────────   │     │
│  │  Tools: datetime · web_search (Tavily) · search_knowledge_base  │     │
│  │  Providers: OpenAI · Anthropic · Gemini · OpenRouter            │     │
│  └─────────────────────────────────────────────────────────────────┘     │
│                                                                          │
│  ┌─────────────────────────────────────────────────────────────────┐     │
│  │                     RAG PIPELINE                                │     │
│  │                                                                 │     │
│  │  Sources        Parse           Chunk          Embed            │     │
│  │  ─────────      ──────────      ──────────     ──────────────   │     │
│  │  Local files    PyMuPDF         recursive      OpenAI           │     │
│  │  API upload     LiteParse       markdown       Voyage           │     │
│  │  Google Drive   LlamaParse      fixed          Gemini (multi)   │     │
│  │  S3/MinIO       python-docx                    SentenceTransf.  │     │
│  │  Sync Sources                                                   │     │
│  │                                                                 │     │
│  │  Store              Search              Rank                    │     │
│  │  ──────────────     ──────────────      ──────────────          │     │
│  │  Milvus             Vector similarity   Cohere reranker         │     │
│  │  Qdrant             BM25 + vector RRF   CrossEncoder            │     │
│  │  ChromaDB           Multi-collection                            │     │
│  │  pgvector                                                       │     │
│  └─────────────────────────────────────────────────────────────────┘     │
│                                                                          │
│  Auth (JWT/API Key/OAuth) · Rate Limiting · Webhooks · Admin Panel       │
│  Billing (Stripe + credits) · Background Tasks (Celery/Taskiq/ARQ/       │
│  Prefect) · Django-style CLI · Observability (Logfire/LangSmith/         │
│  Sentry/Prometheus)                                                      │
└───────┬──────────────┬──────────────┬──────────────┬─────────────────────┘
        │              │              │              │
        ▼              ▼              ▼              ▼
   PostgreSQL       Redis         Vector DB      LLM APIs
   (async)                        (Milvus/       (OpenAI/
                                  Qdrant/        Anthropic/
                                  ChromaDB/      Gemini)
                                  pgvector)

🏗️ Architecture

graph TB
    subgraph Frontend["Frontend (Next.js 15)"]
        UI[React Components]
        WS[WebSocket Client]
        Store[Zustand Stores]
    end

    subgraph Backend["Backend (FastAPI)"]
        API[API Routes]
        Services[Services Layer]
        Repos[Repositories]
        Agent[AI Agent]
    end

    subgraph Infrastructure
        DB[(PostgreSQL)]
        Redis[(Redis)]
        Queue[Celery/Taskiq/ARQ/Prefect]
    end

    subgraph External
        LLM[OpenAI/Anthropic]
        Webhook[Webhook Endpoints]
    end

    UI --> API
    WS <--> Agent
    API --> Services
    Services --> Repos
    Services --> Agent
    Repos --> DB
    Agent --> LLM
    Services --> Redis
    Services --> Queue
    Services --> Webhook

Layered Architecture

The backend follows a clean Repository + Service pattern:

graph LR
    A[API Routes] --> B[Services]
    B --> C[Repositories]
    C --> D[(Database)]

    B --> E[External APIs]
    B --> F[AI Agents]

See Architecture Documentation for details.

🤖 AI Agent

Choose from 5 AI frameworks and 4 LLM providers when generating your project:

# PydanticAI with OpenAI (default)
fastapi-fullstack create my_app --ai-framework pydantic_ai

# LangGraph with Anthropic
fastapi-fullstack create my_app --ai-framework langgraph --llm-provider anthropic

# DeepAgents with OpenAI
fastapi-fullstack create my_app --ai-framework deepagents

# With RAG enabled
fastapi-fullstack create my_app --rag --database postgresql --task-queue celery

Supported Combinations

PydanticAI Integration

Type-safe agents with full dependency injection:

# app/agents/assistant.py
from pydantic_ai import Agent, RunContext

@dataclass
class Deps:
    user_id: str | None = None
    db: AsyncSession | None = None

agent = Agent[Deps, str](
    model="openai:gpt-4o-mini",
    system_prompt="You are a helpful assistant.",
)

@agent.tool
async def search_database(ctx: RunContext[Deps], query: str) -> list[dict]:
    """Search the database for relevant information."""
    # Access user context and database via ctx.deps
    ...

LangChain Integration

Flexible agents with LangGraph:

# app/agents/langchain_assistant.py
from langchain.tools import tool
from langgraph.prebuilt import create_react_agent

@tool
def search_database(query: str) -> list[dict]:
    """Search the database for relevant information."""
    ...

agent = create_react_agent(
    model=ChatOpenAI(model="gpt-4o-mini"),
    tools=[search_database],
    prompt="You are a helpful assistant.",
)

WebSocket Streaming

Both frameworks use the same WebSocket endpoint with real-time streaming:

@router.websocket("/ws")
async def agent_ws(websocket: WebSocket):
    await websocket.accept()

    # Works with both PydanticAI and LangChain
    async for event in agent.stream(user_input):
        await websocket.send_json({
            "type": "text_delta",
            "content": event.content
        })

Observability

Each framework has its own observability solution:

See AI Agent Documentation for more.

📄 RAG (Retrieval-Augmented Generation)

Enable RAG to give your AI agents access to a knowledge base built from your documents.

Vector Store Backends

Document Ingestion (CLI)

# Local files
uv run my_app rag-ingest /path/to/document.pdf --collection docs
uv run my_app rag-ingest /path/to/folder/ --recursive

# Google Drive (service account)
uv run my_app rag-sync-gdrive --collection docs --folder-id <drive_folder_id>

# S3/MinIO
uv run my_app rag-sync-s3 --collection docs --prefix reports/ --bucket my-bucket

Embedding Providers

Features

• Document parsing - PDF (PyMuPDF with tables, headers/footers, OCR), DOCX, TXT, MD + 130+ formats via LlamaParse
• Image description - Extract images from documents, describe via LLM vision API (opt-in)
• Chunking - RecursiveCharacterTextSplitter with configurable size/overlap
• Reranking - Cohere API or local CrossEncoder for improved search quality
• Agent integration - All 5 AI frameworks get a search_knowledge_base tool automatically

📊 Observability

Logfire (for PydanticAI)

Logfire provides complete observability for your application - from AI agents to database queries. Built by the Pydantic team, it offers first-class support for the entire Python ecosystem.

graph LR
    subgraph Your App
        API[FastAPI]
        Agent[PydanticAI]
        DB[(Database)]
        Cache[(Redis)]
        Queue[Celery/Taskiq]
        HTTP[HTTPX]
    end

    subgraph Logfire
        Traces[Traces]
        Metrics[Metrics]
        Logs[Logs]
    end

    API --> Traces
    Agent --> Traces
    DB --> Traces
    Cache --> Traces
    Queue --> Traces
    HTTP --> Traces

LangSmith (for LangChain)

LangSmith provides observability specifically designed for LangChain applications:

LangSmith is automatically configured when you choose LangChain:

# .env
LANGCHAIN_TRACING_V2=true
LANGCHAIN_API_KEY=your-api-key
LANGCHAIN_PROJECT=my_project

Configuration

Enable Logfire and select which components to instrument:

fastapi-fullstack new
# ✓ Enable Logfire observability
#   ✓ Instrument FastAPI
#   ✓ Instrument Database
#   ✓ Instrument Redis
#   ✓ Instrument Celery
#   ✓ Instrument HTTPX

Usage

# Automatic instrumentation in app/main.py
import logfire

logfire.configure()
logfire.instrument_fastapi(app)
logfire.instrument_asyncpg()
logfire.instrument_redis()
logfire.instrument_httpx()

# Manual spans for custom logic
with logfire.span("process_order", order_id=order.id):
    await validate_order(order)
    await charge_payment(order)
    await send_confirmation(order)

For more details, see Logfire Documentation.

🛠️ Django-style CLI

Each generated project includes a powerful CLI inspired by Django's management commands:

Built-in Commands

# Server
my_app server run --reload
my_app server routes

# Database (Alembic wrapper)
my_app db init
my_app db migrate -m "Add users"
my_app db upgrade

# Users
my_app user create --email admin@example.com --superuser
my_app user list

Custom Commands

Create your own commands with auto-discovery:

# app/commands/seed.py
from app.commands import command, success, error
import click

@command("seed", help="Seed database with test data")
@click.option("--count", "-c", default=10, type=int)
@click.option("--dry-run", is_flag=True)
def seed_database(count: int, dry_run: bool):
    """Seed the database with sample data."""
    if dry_run:
        info(f"[DRY RUN] Would create {count} records")
        return

    # Your logic here
    success(f"Created {count} records!")

Commands are automatically discovered from app/commands/ - just create a file and use the @command decorator.

my_app cmd seed --count 100
my_app cmd seed --dry-run

📁 Generated Project Structure

my_project/
├── backend/
│   ├── app/
│   │   ├── main.py              # FastAPI app with lifespan
│   │   ├── api/
│   │   │   ├── routes/v1/       # Versioned API endpoints
│   │   │   ├── deps.py          # Dependency injection
│   │   │   └── router.py        # Route aggregation
│   │   ├── core/                # Config, security, middleware
│   │   ├── db/models/           # SQLAlchemy 2.0 models
│   │   ├── schemas/             # Pydantic schemas
│   │   ├── repositories/        # Data access layer
│   │   ├── services/            # Business logic
│   │   ├── agents/              # AI agents with centralized prompts
│   │   ├── rag/                 # RAG module (vector store, embeddings, ingestion)
│   │   ├── commands/            # Django-style CLI commands
│   │   └── worker/              # Background tasks
│   ├── cli/                     # Project CLI
│   ├── tests/                   # pytest test suite
│   └── alembic/                 # Database migrations
├── frontend/
│   ├── src/
│   │   ├── app/                 # Next.js App Router
│   │   ├── components/          # React components
│   │   ├── hooks/               # useChat, useWebSocket, etc.
│   │   └── stores/              # Zustand state management
│   └── e2e/                     # Playwright tests
├── docker-compose.yml
├── Makefile
└── README.md

Generated projects include version metadata in pyproject.toml for tracking:

[tool.fastapi-fullstack]
generator_version = "0.1.5"
generated_at = "2024-12-21T10:30:00+00:00"

⚙️ Configuration Options

Core Options

Presets

Integrations

Select what you need:

fastapi-fullstack new
# ✓ Redis (caching/sessions)
# ✓ Rate limiting (slowapi)
# ✓ Pagination (fastapi-pagination)
# ✓ Admin Panel (SQLAdmin)
# ✓ AI Agent (PydanticAI or LangChain)
# ✓ Webhooks
# ✓ Sentry
# ✓ Logfire / LangSmith
# ✓ Prometheus
# ... and more

🔄 Comparison

vs. Manual Setup

Setting up a production AI agent stack manually means wiring together 10+ tools yourself:

# Without this template, you'd need to manually:
# 1. Set up FastAPI project structure
# 2. Configure SQLAlchemy + Alembic migrations
# 3. Implement JWT auth with refresh tokens
# 4. Build WebSocket streaming for AI responses
# 5. Integrate PydanticAI/LangChain with tool calling
# 6. Set up RAG pipeline (parsing, chunking, embedding, vector store)
# 7. Configure Celery + Redis for background tasks
# 8. Build Next.js frontend with auth and chat UI
# 9. Write Docker Compose for all services
# 10. Add observability, rate limiting, admin panel...

# With this template:
pip install fastapi-fullstack
fastapi-fullstack
# Done. All of the above, configured and working.

vs. Alternatives

❓ FAQ

📚 Documentation

Star History

🙏 Inspiration

This project is inspired by:

• full-stack-fastapi-template by @tiangolo
• fastapi-template by @s3rius
• FastAPI Best Practices by @zhanymkanov
• Django's management commands system

🤝 Contributing

Contributions are welcome! Please read our Contributing Guide for details.

📄 License

MIT License - see LICENSE for details.

Need help implementing this in your company?

We're Vstorm — an Applied Agentic AI Engineering Consultancy
with 30+ production AI agent implementations.

Made with ❤️ by Vstorm