fast-agent-stack

Production infrastructure for AI agent applications, built on FastAPI.

fast-agent-stack gives you auth, database, vector search, background tasks, rate limiting, observability, and storage out of the box. Bring your own agent framework (Strands, Pydantic AI, LangGraph) or use the built-in @app.agent() for simple cases.

---

Quick Start

mkdir myproject &&cd myproject
uv venv &&source .venv/bin/activate
uv pip install fast-agent-stack
fas new myproject --preset agent
uv pip install -r pyproject.toml
fas migrate
fas dev

Visit http://localhost:8000/docs to see the interactive API.

---

Installation

# Core only
pip install fast-agent-stack
# With extras (mix and match)
pip install "fast-agent-stack[auth-jwt,db-postgres,vector-qdrant]"# Full AI stack
pip install "fast-agent-stack[ai-full]"

---

Presets

Pick a preset and get a production-ready project in seconds:

Preset	What you get
minimal	FastAPI app, SQLAlchemy, health checks, CLI
standard	+ JWT auth, SQLAdmin, Docker
full	+ rate limiting, background tasks, email, observability
agent	+ LLM backends, vector store, RAG pipeline, streaming

fas new myproject --preset standard

---

Features

Infrastructure

• SQLAlchemy async + Alembic migrations
• Redis/Valkey: auth, rate limiting, response caching
• Background tasks (Dramatiq) + scheduler (Periodiq)
• Storage: S3, MinIO, local filesystem
• OpenTelemetry tracing (Jaeger backend)
• AWS / GCP secrets managers

Auth

• JWT + session auth backends, pluggable via settings
• RBAC: users, groups, permissions, API keys
• Redis JTI denylist, email verification, password reset
• SQLAdmin UI (optional)

Vector Search & RAG

• Vector stores: Qdrant, pgvector, OpenSearch, Weaviate
• Embedding backends: Bedrock, OpenAI, fastembed (local)
• RAG pipeline: chunk, embed, store / retrieve
• Document extraction: PDF, DOCX, XLSX, EML

AI / Agents

• Built-in @app.agent() decorator for simple single-agent endpoints
• LLM backends: AWS Bedrock, OpenAI, Anthropic, LiteLLM proxy
• get_llm(settings) factory for one-line backend resolution
• Framework integration guides for Strands Agents and Pydantic AI

---

Bring Your Own Agent Framework

fast-agent-stack's built-in @app.agent() covers simple cases (like FastAPI's BackgroundTasks). For serious agentic work, bring your own framework and use fast-agent-stack for the infrastructure:

# myproject/ai/agents/chat.pyfromstrandsimportAgentfromstrands.models.litellmimportLiteLLMModelfrommyproject.ai.tools.searchimportsearch_docsfrommyproject.settingsimportget_settings_settings=get_settings()
defbuild_chat_agent():
model=LiteLLMModel(
model_id=f"openai/{_settings.llm_model}",
params={"api_key": _settings.llm_api_key, "api_base": _settings.llm_base_url},
)
returnAgent(model=model, tools=[search_docs])

# myproject/routes.pyfromfastapi.responsesimportStreamingResponse@router.post("/agents/chat")asyncdefchat(body: ChatRequest) ->StreamingResponse:
returnStreamingResponse(stream_chat(body.message), media_type="text/event-stream")

See the Strands Agents guide or Pydantic AI guide for full working examples.

---

Extras

# Authfast-agent-stack[auth-jwt] # JWT backendfast-agent-stack[auth-session] # Session backend# Database driversfast-agent-stack[db-postgres] # asyncpgfast-agent-stack[db-sqlite] # aiosqlitefast-agent-stack[db-mysql] # aiomysql# LLM backendsfast-agent-stack[anthropic] # Anthropic SDKfast-agent-stack[openai] # OpenAI SDKfast-agent-stack[bedrock] # aioboto3 (AWS Bedrock)fast-agent-stack[litellm] # LiteLLM proxy# Vector storesfast-agent-stack[vector-qdrant]fast-agent-stack[vector-pgvector]fast-agent-stack[vector-opensearch]fast-agent-stack[vector-weaviate]# Background tasksfast-agent-stack[tasks] # Dramatiq + Redis broker# Observabilityfast-agent-stack[tracing] # OpenTelemetry + Jaeger# Full AI bundlefast-agent-stack[ai-full]

---

CLI

Command	Description
fas new <name>	Scaffold a new project
fas dev	Dev server (127.0.0.1, auto-reload)
fas run	Production server (0.0.0.0, multi-worker)
fas migrate	Apply all migrations
fas makemigrations	Generate migration from model changes
fas worker <module>	Start Dramatiq worker
fas scheduler <module>	Start Periodiq scheduler
fas createsuperuser	Create a superuser account
fas version	Print installed version

fastagentstack also works as the full-length alias.

---

Documentation

• Tutorial - Build a Document Q&A app step by step
• Framework Integration Guides - Strands Agents, Pydantic AI
• Getting Started
• Authentication & Authorization
• AI Module & Agents
• RAG Pipeline
• Storage & Extraction
• Background Tasks & Scheduling
• Rate Limiting & Observability
• Deployment
• Custom Backends
• Configuration Reference

---

License

MIT