systemgen

CLI tool that builds reusable AI context from your stack and turns it into a single prompt you can send to an assistant.

Building with AI works better when the model knows your framework, database, folder layout, and conventions. Re-typing that context for every task wastes time and drifts from how your repo actually works. systemgen captures those choices once as a named profile, then merges them with whatever you want to build next. Profiles can live as TOML (terminal tools) or JSON (IDE assistants), or both.

Install

pip install systemgen

Quick start

systemgen init
# Answer prompts for stack, structure, and conventions; pick TOML, JSON, or both; name the profile.

systemgen gen --profile myapp
# Pick your profile if you omit --profile; choose how to supply the task prompt; then send the result where you need it.

Commands

Command	What it does
systemgen init	Create a new profile from the full question flow.
systemgen gen	Build a prompt from a saved profile (--profile <name> optional).
systemgen update	Show current profile values, then full re-run of all questions, field-by-field edits, or cancel (--profile <name> optional); same name on save.
systemgen profile list	List saved profiles and optional follow-up actions.
systemgen profile delete	Remove a profile (TOML, JSON, or both).
systemgen profile library	Import from built-in templates or start from scratch.

Run systemgen with no subcommand to open the interactive menu when you have profiles.

Profile formats

Profiles are plain data. TOML suits terminal workflows (e.g. Claude Code, Ollama); JSON suits IDE tooling (e.g. Cursor, Windsurf). Example shape (FastAPI-style), shortened:

TOML (~/.systemgen/profiles/<name>.toml):

[project]
type = "Backend API"
framework = "FastAPI"
language = "python"
language_version = "3.11"

[database]
engine = "PostgreSQL"
orm = "sqlalchemy"
mode = "async"

[auth]
method = "JWT"

JSON (~/.systemgen/profiles/<name>.json):

{
  "project": {
    "type": "Backend API",
    "framework": "FastAPI",
    "language": "python",
    "language_version": "3.11"
  },
  "database": {
    "engine": "PostgreSQL",
    "orm": "sqlalchemy",
    "mode": "async"
  },
  "auth": {
    "method": "JWT"
  }
}

Prompt input options (gen)

After you pick a profile, you choose how to add the task:

Option	When to use it
Type it now	You know the request and want to enter it in the terminal.
Load from a text file	The spec lives in a .txt or .md file (paths are re-checked until the file exists).
Skip — I'll write the prompt in my AI tool	You only want the profile context; you will add the task inside the assistant.

If you skip the task text, the next step only offers copy, save to file, or exit (no pipe to Claude/Ollama).

Pipe to (gen)

When you supplied a task prompt, you can send the combined output to:

Destination	What it does
Claude (terminal)	Starts the Claude Code CLI with your prompt as the opening message so you can keep chatting.
Ollama	Runs ollama run <model> with the full text as input (model list comes from ollama list when available).
Copy to clipboard	Copies the full prompt for pasting anywhere.
Save to file	Writes the prompt to a path you enter.

Profiles are saved to

~/.systemgen/profiles/

Languages and stacks

systemgen is not limited to Python. You pick language (e.g. TypeScript, Dart, Go), framework, database, auth, validation, and folder layout from prompts or templates—use it for APIs, frontends, mobile, or CLIs on whatever stack you use.