active

AI Ansible Generator

Plain English → validated Ansible playbooks

Turns a plain-English request into an Ansible playbook, then validates YAML, syntax and safety and runs a repair loop until it passes. FastAPI + Gradio, backed by Ollama or Hugging Face models.

Python
FastAPI
Gradio
Ollama
Hugging Face
Ansible

Live demo repository is private

01 problem

Large language models can write Ansible, but raw model output can't be trusted in production: it may not be valid YAML, may not pass Ansible's own syntax checks, and may do something destructive. Pasting it into a terminal is how outages start.

02 approach

01 Treat the model as an untrusted author: everything it writes goes through validation before anyone sees it as a result.
02 Validate in layers: YAML parsing, Ansible syntax, then safety rules for risky patterns.
03 When validation fails, feed the errors back to the model in a repair loop instead of handing the user a broken playbook.
04 Keep the model pluggable: run locally with Ollama or use hosted models through Hugging Face.
05 Serve it two ways: a FastAPI backend for programmatic use and a Gradio interface for people.

03 architecture

AI Ansible Generator — generate, validate, repair select a step

01 · request
02 · generate
03 · validate
04 · repair
05 · result

trigger · gradio ui · fastapi

plain-English request

A person describes the automation they want in plain English through the Gradio interface, or a client calls the FastAPI backend.

All steps as text

1. request
- plain-English request: A person describes the automation they want in plain English through the Gradio interface, or a client calls the FastAPI backend.
2. generate
- LLM: The request is turned into a draft playbook by a local model through Ollama or a hosted model through Hugging Face. Either way, the draft is treated as untrusted.
3. validate
- YAML check: The draft must parse as YAML before anything else happens.
- syntax check: The playbook must pass Ansible syntax validation.
- safety rules: Safety checks flag dangerous operations before the playbook is returned.
4. repair
- repair loop: If any check fails, the validation errors go back to the model with the draft so it can fix them, and the result is validated again.
5. result
- validated playbook: Only a playbook that passed every check is returned to the user or API client.

04 outcome

Plain-English requests become Ansible playbooks that have passed YAML, syntax and safety validation.
A live demo runs on Hugging Face Spaces.