LLM Automation Agent

Execute plain-English tasks via an LLM-powered agent. No proxy — OpenAI, Gemini, or OpenRouter only.

Built with

---

Overview

The LLM Automation Agent accepts plain-English task descriptions over an API, dispatches them to a tool-calling agent (LangChain or Pydantic AI), and returns results. Tools cover SQL (SQLite/DuckDB), OCR, image extraction, audio transcription, embeddings, web scraping, Markdown/HTML, CSV filtering, and more. All paths are restricted to /data; no file deletion. LLM calls use OpenAI, Gemini, or OpenRouter — no proxy.

Features

• POST /run?task=<plain-English> — Run the agent on a task. Returns { "status": "success", "message": "..." } or 500 on error.
• GET /read?path=<file path> — Read a file under /data to verify outputs (403 if outside /data, 404 if missing).
• Path guards — All file I/O restricted to /data via ensure_local_path / enforce_data_directory.
• Dual backends — LangChain (default) or Pydantic AI via AGENT_BACKEND.
• Optional GPTCache — Cache LLM responses when GPT_CACHE_ENABLED=true.
• Docker — Dockerfile and Podman-compatible; run with OPENAI_API_KEY or OPENROUTER_API_KEY.

Quick start

git clone https://github.com/your-username/llm-proj-final.git
cd llm-proj-final
pip install uv
uv venv
source venv/bin/activate   # Windows: .\venv\Scripts\activate
uv pip install -r requirements.txt -r requirements-dev.txt

Set OPENAI_API_KEY or OPENROUTER_API_KEY (and LITELLM_MODEL if using OpenRouter). Use a .env in the project root or in app/.

cd app && uv run uvicorn main:app --reload --host 0.0.0.0 --port 8000

• API: http://localhost:8000
• Docs: http://localhost:8000/docs

Configuration

Variable	Purpose
OPENAI_API_KEY	OpenAI API key (required for default OpenAI model).
OPENAI_API_BASE	OpenAI API base URL (default: https://api.openai.com/v1).
GOOGLE_API_KEY	Google API key when using Gemini models via LiteLLM.
LITELLM_MODEL	Model string, e.g. openai/gpt-4o-mini (default) or openrouter/openai/gpt-4o-mini.
OPENROUTER_API_KEY	Required when LITELLM_MODEL uses openrouter/....
LANGCHAIN_TRACING_V2	Set to true to enable LangSmith tracing.
LANGCHAIN_API_KEY	LangSmith API key (when tracing enabled).
LANGCHAIN_PROJECT	LangSmith project name (default: llm-proj-final).
GPT_CACHE_ENABLED	Set to true to enable GPTCache for LangChain LLM calls.
AGENT_BACKEND	langchain (default) or pydantic_ai.

Docker

docker build -t llm-proj-final .
docker run --rm -e OPENAI_API_KEY=$OPENAI_API_KEY -p 8000:8000 llm-proj-final

With OpenRouter:

docker run --rm -e OPENROUTER_API_KEY=$OPENROUTER_API_KEY -e LITELLM_MODEL=openrouter/openai/gpt-4o-mini -p 8000:8000 llm-proj-final

Podman:

podman run --rm -e OPENAI_API_KEY=$OPENAI_API_KEY -p 8000:8000 llm-proj-final

API

Endpoint	Description
POST /run?task=<plain-English>	Run the agent on a task. Returns { "status": "success", "message": "..." } or 500 on error.
GET /read?path=<file path>	Read a file under /data. Returns 403 if path is outside /data, 404 if not found.

CI

GitHub Actions runs on push and pull requests: Ruff (check + format), pytest, and Docker build. The agent smoke test uses OpenRouter; add an OPENROUTER_API_KEY repository secret (or use OPENAI_API_KEY with OpenAI) to run it. If neither is set, the smoke test is skipped.

See .github/workflows/ci.yml.

Project structure

llm-proj-final/
├── app/
│   ├── main.py           # FastAPI app; /run, /read
│   ├── config.py         # Env-based config (LiteLLM, LangSmith, GPTCache, AGENT_BACKEND)
│   ├── funtion_tasks.py  # Tool implementations (path guards, no deletion)
│   ├── agent_langchain.py
│   ├── agent_pydantic_ai.py
│   └── task_to_embed.txt
├── tests/
│   └── test_api.py
├── .github/workflows/
│   └── ci.yml
├── Dockerfile
├── pyproject.toml
├── requirements.txt
├── requirements-dev.txt
└── README.md

Contributing

1. Fork the repository.
2. Create a feature branch.
3. Commit changes with meaningful messages.
4. Open a pull request.

License & credits

This project is licensed under the MIT License. See LICENSE for details.

• Author: Sandeep S