AI-Powered Coder Workspace Template Generator Auto-detects local LLMs, Google Gemini, Claude, OpenRouter — generates production-grade Terraform in seconds.

TerraForge takes any description of a dev environment and generates a complete, valid Coder workspace template — HCL, README, tfvars, the works.

It auto-detects every LLM available on your machine and in your environment, picks the best one, and streams the generation live in your terminal or web UI.

You:         "python fastapi workspace with postgres on docker, medium size"
TerraForge:  ✓ Detected Gemini → gemini-1.5-flash
             [streams generated HCL live]
             ✓ Validation passed
             ✓ Saved: output/main.tf, output/README.md, output/terraform.tfvars.example

TerraForge scans all sources in parallel and picks the best one automatically:

Local LLMs always win. If Ollama is running, it's used. If not, Gemini is the default cloud provider (free, no card required for 1,500 req/day).

TerraForge uses Gemini 1.5 Flash as its default cloud provider:

• Free tier: 1,500 requests/day — no credit card required
• Get a key: aistudio.google.com/apikey
• Set it: export GOOGLE_API_KEY=AIza… or enter it in the web UI Settings panel

Gemini is placed at priority 15 — it beats paid cloud providers but yields to any running local model.

git clone <this-repo>
cd terraforge
bash setup.sh
source .venv/bin/activate

setup.sh installs dependencies, scans for LLMs, and prints a tip if GOOGLE_API_KEY is not set.

python terraforge.py "rust actix-web workspace with redis, large size, on docker"
python terraforge.py "go microservices workspace for kubernetes with kubectl helm k9s"
python terraforge.py "python ML workspace with pytorch and jupyter, GPU, on AWS EC2"

python terraforge.py examples/python-ml.yaml
python terraforge.py examples/go-microservices.json
python terraforge.py examples/fullstack-nextjs.md

python terraforge.py --interactive

python terraforge.py --detect

python terraforge.py examples/python-ml.yaml --push
# Requires: coder login (or CODER_URL + CODER_SESSION_TOKEN)

python terraforge.py --detect            # shows numbered list
python terraforge.py "..." --provider 2  # use provider #2

Launch the full web UI (generator + settings panel):

python terraforge.py server
# → http://localhost:7842/app

The web UI provides:

• Generator — same as CLI, with live streaming output and HCL syntax highlighting
• Settings panel — configure API keys and Coder connection without touching files
• Provider sidebar — see all detected providers, select and switch models
• Download — save generated files locally
• Push to Coder — deploy directly from the browser

Click the ⚙ Settings button in the top-right to open the settings panel:

Keys are saved to ~/.terraforge/config.json with permissions 600.

name: my-workspace
display_name: My Dev Workspace
description: Python backend workspace
target: docker          # docker | aws_ec2 | kubernetes | gcp_gke | azure_aks | hetzner | digitalocean
language: python
ide: code-server        # code-server | jetbrains | jupyter

compute:
  cpu: 4
  memory_gb: 8
  disk_gb: 50
  gpu: false

software:
  version: "3.11"
  frameworks: [fastapi, sqlalchemy]
  tools: [git, curl, postgresql-client]

cost:
  auto_stop_hours: 2
  auto_delete_days: 7

{
  "name": "go-api",
  "target": "kubernetes",
  "language": "go",
  "compute": { "cpu": 4, "memory_gb": 8, "disk_gb": 50 }
}

Any .md file with optional YAML frontmatter. The body becomes additional context for the LLM.

API keys are never committed to this repository and never sent to TerraForge servers.

Sensitive patterns protected by .gitignore:

• .env / .env.*
• *.tfvars (except *.tfvars.example)
• terraform.tfstate / .terraform/
• config.json

If you accidentally commit a secret, rotate the key immediately.

output/
├── main.tf                    # Complete Terraform template
├── README.md                  # Auto-generated docs
└── terraform.tfvars.example   # Configurable defaults

Push to Coder:

coder templates push my-workspace --directory output/

• Python 3.11+
• At least one LLM source (Gemini free tier works with just a key)

Optional (enhance experience):

• terraform — enables HCL formatting & validation
• coder CLI — enables --push to deploy directly

1. qwen2.5-coder:32b / 14b / 7b — best for code generation
2. deepseek-coder-v2:16b
3. codellama:34b / 13b
4. llama3.1:70b / 8b
5. Any other available model

Open this folder in VS Code — everything is pre-configured:

• Ctrl+Shift+P → Tasks: Run Task — all TerraForge commands as tasks
• Run & Debug panel — pre-built launch configurations
• Recommended extensions auto-suggested on open

TerraForge — Stop writing Terraform by hand.