AGX: The Verifiable AI Workflow Engine for DevOps

Verification First Terraform

AGX is a single-shot agentic Terraform generator that constrains the LLM planner to a predefined function registry, statically validating each generated plan before compilation. It checks for function existence, parameter usage, variable assignment order, and type correctness.

An independent project exploring whether moving validation earlier and restricting output to a known function registry can produce more predictable LLM-generated infrastructure.

GitHub

Describe your infrastructure...

Python to TF Script

About the Engine

How it works:

Uses a deterministic validation stage to enforce strict type safety. The engine checks every step against a function registry to prevent hallucinated parameters or invalid dependencies.

Future work:

The most significant missing piece is DAG-based validation — currently the validator only checks plan format, not execution order or dependency soundness.

Available tools (demo subset):

  • set_bucket_name
  • create_aws_s3_bucket
  • aws_s3_bucket_public_access_block
  • save_hcl_to_file
  • sanitise_resource_name
  • combine_two_hcl_blocks
Built by Ammar Qureshi