Type your feature idea in plain English. Get a professional use case document back.

ucgen is a Python CLI that generates Cockburn-style use case documents from natural language using Ollama or hosted providers.

Install

pip install ucgen

Quick Start

Local (Ollama)

ollama pull qwen3:8b
ucgen generate "warehouse picker scans barcode before adding item to shipment"

Cloud (Anthropic)

export ANTHROPIC_API_KEY=sk-ant-...   # Windows: $env:ANTHROPIC_API_KEY=...
ucgen generate "warehouse picker scans barcode before adding item to shipment" --provider anthropic --model claude-sonnet-4-6

Core Concepts

• One idea -> one use case by default (ucgen generate).
• --expand asks AI to discover multiple use cases first, then generates them sequentially.
• Output is always grounded in a fixed 4-stage generation pipeline.

What ucgen Writes

A successful run creates a standalone folder per use case:

use-cases/
  standalone/
    UC-2026-0049-warehouse-picker-scans-barcode/
      UC-2026-0049-warehouse-picker-scans-barcode.md
      UC-2026-0049-warehouse-picker-scans-barcode.json
      report.html                # optional: with --report
      UC-2026-0049-... .docx     # optional: with --docx / --claude

• .md: human-readable use case document
• .json: structured UseCaseDocument output
• .docx: optional Word output
• report.html: optional HTML report

Word Document Generation

You can generate Word documents in two ways after markdown is written:

Native python-docx export

ucgen generate "register a new patient at reception" --docx

Claude Code terminal flow

ucgen generate "register a new patient at reception" --claude

This launches Claude Code to generate a .docx from the produced markdown.

Requirements:

npm install -g @anthropic-ai/claude-code

Both flags can be combined. --docx runs first, then --claude:

ucgen generate "register a new patient at reception" --docx --claude

Multi-Use-Case Discovery (--expand)

Use this when your idea describes a whole product/workflow, not just one use case:

ucgen generate "A CCTV desktop app monitor for a farm" --expand

Flow:

1. AI proposes all relevant use cases in a table
2. You confirm generation
3. ucgen runs the full 4-stage pipeline for each discovered use case
4. Final panel summarizes generated IDs/titles

Useful Flags

• --report: write a single-use-case HTML report in the output folder
• --quiet: print only generated markdown path(s), ideal for CI/pipes
• --no-color: disable ANSI color output
• --provider, --model: override provider/model per run
• --stdin: read idea from stdin instead of CLI argument

Example Output Excerpt

Register New Patient (UC-2026-0049)

---
uc_id: UC-2026-0049
title: Register New Patient
actor: Receptionist
goal_level: user_goal
---

## Preconditions
- Patient identification documents are available
- Registration desk is staffed

## Normal Course
| Step | Actor | Action | System Response |
|------|-------|--------|-----------------|
| 1 | Receptionist | Capture demographics | Validate required fields |
| 2 | Receptionist | Submit registration | Create patient record and assign MRN |

## Success Guarantee
The patient has an active record and can be scheduled for care.

Configuration

Create local config:

ucgen init

Example .ucgenrc.toml:

[defaults]
provider = "ollama"
model = "qwen3:8b"
output_dir = "./use-cases"
template = "default"
temperature = 0.3
max_tokens = 4000

Command Reference

ucgen generate

Generate one use case from a plain-English idea.

ucgen generate "<idea>"

Common options:

• --expand: discover and generate multiple use cases
• --report: generate single-use-case HTML report
• --docx: export native Word file via python-docx
• --claude: run Claude Code Word generation flow
• --quiet: output path(s) only
• --provider, --model: provider/model override
• --stdin: read idea from stdin
• --debug: persist debug artifacts on parse failures

ucgen batch

Generate multiple use cases from a .txt file (one idea per line) or .yaml list.

ucgen batch ideas.txt
ucgen batch ideas.yaml --report

ucgen run

Generate pending use cases from a ucgen.yaml project file.

ucgen run
ucgen run --id UC-001
ucgen run --tag booking

ucgen status

Show generation status for use cases in a ucgen.yaml project.

ucgen status

ucgen report

Build a portfolio HTML report from all generated JSON files.

ucgen report
ucgen report --output ./use-cases/report.html --open

ucgen validate

Validate generated markdown files against expected structure.

ucgen validate ./use-cases/standalone

ucgen init

Create a .ucgenrc.toml config in the current directory.

ucgen init

ucgen init-project

Scaffold a new ucgen.yaml project definition.

ucgen init-project "Farm CCTV"

ucgen log

Log a mistake/correction entry to ucgen memory.

ucgen log --quick "Used wrong provider default in CLI output"

ucgen gc

Archive or graduate memory entries based on recurrence/staleness rules.

ucgen gc
ucgen gc --dry-run

ucgen version

Print version and provider availability checks.

ucgen version

Troubleshooting

Ollama not reachable
Run ollama serve, then verify with ollama list.

Claude Code not found
Install with npm install -g @anthropic-ai/claude-code.

python-docx missing
Install with pip install python-docx.

Stage 2 retry message
Expected with some models (for example Mistral). Retry auto-corrects structure.

Slow generation
Try --provider groq for faster cloud inference.

License

MIT — see LICENSE.