Single File Agents

Single File Agents (SFA)

Premise: #1: What if we could pack single purpose, powerful AI Agents into a single python file?

Premise: #2: What's the best structural pattern for building Agents that can improve in capability as compute and intelligence increases?



What is this?

A collection of powerful single-file agents built on top of [uv](https://github.com/astral/uv) - the modern Python package installer and resolver.

These agents aim to do one thing and one thing only. They demonstrate precise prompt engineering and GenAI patterns for practical tasks many of which I share on the [IndyDevDan YouTube channel](https://www.youtube.com/@indydevdan). Watch us walk through the Single File Agent in [this video](https://youtu.be/YAIJV48QlXc).

You can also check out [this video](https://youtu.be/vq-vTsbSSZ0) where we use [Devin](https://devin.ai/), [Cursor](https://www.cursor.com/), [Aider](https://aider.chat/), and [PAIC-Patterns](https://agenticengineer.com/principled-ai-coding) to build three new agents with powerful spec (plan) prompts.

This repo contains a few agents built across the big 3 GenAI providers (Gemini, OpenAI, Anthropic).

Quick Start

Export your API keys:

export GEMINI_API_KEY='your-api-key-here'

export OPENAI_API_KEY='your-api-key-here'

export ANTHROPIC_API_KEY='your-api-key-here'

export FIRECRAWL_API_KEY='your-api-key-here' # Get your API key from https://www.firecrawl.dev/

JQ Agent:

uv run sfa_jq_gemini_v1.py --exe "Filter scores above 80 from data/analytics.json and save to high_scores.json"

DuckDB Agent (OpenAI):

# Tip tier
uv run sfa_duckdb_openai_v2.py -d ./data/analytics.db -p "Show me all users with score above 80"

DuckDB Agent (Anthropic):

# Tip tier
uv run sfa_duckdb_anthropic_v2.py -d ./data/analytics.db -p "Show me al