ACF - AgentCodeFactory Local Edition

AI-powered code generation pipeline. Run locally or use cloud APIs.

Read the Manifesto — Why we believe local AI is the future of secure, private code generation.

Features

• Flexible Backends: Ollama, LM Studio (local) or OpenAI, Anthropic (cloud)
• 7-Stage Pipeline: SPEC → CONTEXT → DESIGN → IMPLEMENTATION → TESTING → REVIEW → DONE
• Multi-Model Routing: Automatically routes tasks to appropriate model sizes
• Extension Marketplace: Install premium agents, profiles, and RAG kits
• Local Git Versioning: Every iteration auto-committed for easy rollback

Quick Start

1. Install

git clone https://github.com/Tennisee-data/acf.git
cd acf
pip install -e .

2. Start your LLM backend (required before running)

Pick one:

# Option A: Ollama (recommended - local, free)
brew install ollama          # macOS (see ollama.ai for Linux/Windows)
ollama serve                 # Start server (keep this running!)
ollama pull qwen2.5-coder:7b # In another terminal

# Option B: LM Studio (local, GUI)
# Download from https://lmstudio.ai, start local server on port 1234

# Option C: Cloud API
export OPENAI_API_KEY=sk-...      # or
export ANTHROPIC_API_KEY=sk-...

3. Verify and run

acf check                              # Verify LLM backend is working
acf run "Add user authentication"      # Run your first feature

Important: The pipeline will fail if no LLM backend is available. Make sure Ollama is running (ollama serve) or LM Studio's server is started before running acf run.

Detailed setup guides:

• Ollama Setup Guide - Recommended for CLI users
• LM Studio Setup Guide - GUI alternative

Installation

From PyPI (when published)

pip install acf

From Source (local development)

git clone https://github.com/Tennisee-data/acf.git
cd acf
pip install -e .

Note: If you cloned the repo, you must use pip install -e . from the project directory. Running pip install acf will attempt to install from PyPI, not your local copy.

Optional Dependencies

From source:

pip install -e ".[semantic]"   # Semantic RAG (better code retrieval)
pip install -e ".[openai]"     # OpenAI backend
pip install -e ".[anthropic]"  # Anthropic backend
pip install -e ".[full]"       # Everything

From PyPI (when published):

pip install "acf[semantic]"
pip install "acf[openai]"
pip install "acf[anthropic]"
pip install "acf[full]"

Usage

Basic Generation

acf run "Build a REST API with FastAPI"
acf run "Add user authentication" --repo ./my-project
acf run "Build an API" --output ./my-api    # Custom output directory

Each run creates a project directory with code at the root and history in .acf/:

2026-01-24-130821/                # Project directory (run ID by default)
├── app/                          # Your generated code at root level
│   └── main.py
├── requirements.txt
├── .gitignore                    # Includes .acf/
└── .acf/                         # Pipeline history (gitignored)
    └── runs/
        └── 2026-01-24-130821/    # Run artifacts
            ├── state.json        # Pipeline state
            ├── feature_spec.json # Parsed requirements
            ├── design_proposal.json
            ├── change_set.json
            ├── diff.patch
            └── *.md              # Reports

Use --output to specify a custom project directory:

acf run "Build REST API" --output ./my-project

CLI Reference

acf run "feature" [OPTIONS]

Option	Short	Description
--repo PATH	-r	Target repository (default: current directory)
--output PATH	-o	Output directory for generated project (default: ./{run_id}/)
--profile NAME	-p	Configuration profile (default: dev)
--auto-approve	-y	Skip all approval prompts
--resume ID		Resume a paused run by ID
--dry-run		Show what would be done without executing

Pipeline Stages (built-in):

Option	Description
--decompose	Break feature into sub-tasks before design
--api-contract	Generate OpenAPI spec before implementation
--coverage	Enforce test coverage (default: 80%)
--coverage-threshold N	Set coverage threshold (e.g., 90)
--secrets-scan	Detect hardcoded secrets, auto-fix with env vars
--dependency-audit	Scan for CVEs and outdated packages
--rollback-strategy	Generate CI/CD rollback and canary templates
--observability	Inject logging, metrics, tracing scaffolding
--docs	Generate documentation and ADRs
--code-review	Senior engineer code review
--policy	Enforce policy rules before verification
--pr-package	Build PR with changelog and spec links

Marketplace Extensions (install separately):

Option	Description
--config	Enforce 12-factor config layout

Issue Integration:

Option	Description
--jira PROJ-123	Fetch requirements from JIRA issue
--issue URL	Fetch from GitHub/JIRA issue URL

Examples

# Basic run with auto-approval
acf run "Add login rate-limit" --auto-approve

# Full pipeline with all quality checks
acf run "Build payment API" --api-contract --coverage --secrets-scan --code-review

# Resume a paused run
acf run "Add feature" --resume 2026-01-24-130821

# Iterate on existing project
acf iterate 2026-01-24-130821 "Add error handling"

Other Commands

# List all runs
acf list

# Show run details
acf show 2026-01-24-130821

# Extract generated code to a directory
acf extract 2026-01-24-130821 --output ./my-project

# Create git scaffold with proper structure
acf scaffold 2026-01-24-130821

# Generate tests for existing code
acf generate-tests 2026-01-24-130821

# Deploy generated project
acf deploy 2026-01-24-130821 --version v1.0.0

Extension Marketplace

# Browse extensions
acf marketplace search "security"
acf marketplace featured

# Install extensions
acf marketplace install secrets-scan
acf marketplace install semantic-rag

# Manage installed extensions
acf extensions list
acf extensions enable secrets-scan

Skills (Standalone Code Transformations)

Skills are standalone, composable code transformations that run directly on files without requiring a full pipeline execution.

# Run a skill on a file or directory
acf skill run add-error-handling ./src/main.py
acf skill run add-type-hints ./src --dry-run

# Run a chained skill (multi-step transformation)
acf skill chain production-ready ./src

# List and inspect installed skills
acf skill list
acf skill show add-error-handling

Creating a skill:

1. Create a directory with manifest.yaml and skill.py
2. Install: cp -r my-skill ~/.coding-factory/extensions/skills/

# manifest.yaml
name: my-skill
version: 1.0.0
type: skill
author: Your Name
description: What it does
license: free
skill_class: MySkill
input_type: files
output_type: modified_files
file_patterns: ["*.py"]
supports_dry_run: true

# skill.py
from skills.base import BaseSkill, FileChange, SkillInput, SkillOutput

class MySkill(BaseSkill):
    def run(self, input_data: SkillInput) -> SkillOutput:
        changes = []
        for path in input_data.target_paths:
            content = path.read_text()
            modified = self.transform(content)
            if modified != content:
                changes.append(FileChange(
                    path=path,
                    original_content=content,
                    modified_content=modified,
                    change_type="modified",
                ))
        return SkillOutput(
            success=True,
            changes=changes,
            summary=f"Transformed {len(changes)} files",
        )

    def transform(self, content: str) -> str:
        # Your transformation logic
        return content

Configuration

Create config.toml in your project. Make sure models match what you have installed (ollama list):

[llm]
backend = "auto"  # ollama, lmstudio, openai, anthropic
model_general = "qwen2.5-coder:7b"  # Must match an installed model
model_code = "qwen2.5-coder:7b"

[extensions]
extensions_dir = "~/.coding-factory/extensions"

[routing]
# Enable to use different models for different task complexities
# All models must be installed: ollama pull <model>
enabled = true
model_cheap = "qwen2.5-coder:7b"    # Simple tasks
model_medium = "yi-coder:9b"         # Medium complexity
model_premium = "qwen2.5-coder:32b"  # Complex tasks

Common error: 404 Not Found means the model in config.toml is not installed. Run ollama list to see available models.

Optional Configuration Files

These files are optional but provide additional customization when populated:

File	Purpose
context_libraries.cfg	Allowlist of libraries to include in context analysis
safety_patterns.toml	Custom security patterns for code scanning
tailwind_patterns.toml	Tailwind CSS patterns for frontend generation

Empty template files are included in the repo. Populate them as needed:

context_libraries.cfg - List libraries to always include in context:

# One library per line
requests
fastapi
sqlalchemy

safety_patterns.toml - Define security rules:

[patterns]
# Add custom regex patterns to flag
api_key_in_code = "api[_-]?key\\s*=\\s*['\"][^'\"]+['\"]"

tailwind_patterns.toml - Define Tailwind conventions:

[components]
# Map component types to Tailwind classes
button_primary = "bg-blue-500 hover:bg-blue-600 text-white px-4 py-2 rounded"

---

Marketplace: Build & Sell Extensions

The ACF Marketplace lets you create and sell extensions. Earn money by building agents, profiles, or RAG kits that solve real problems.

Browse the Marketplace | Full Developer Guide

Extension Types

Type	Purpose	Price Range
Agents	Add pipeline stages (security scanning, code review, etc.)	Free - $49
Profiles	Framework templates (Vue, Go, Flutter, etc.)	Free - $15
RAG Kits	Custom code retrieval systems	Free - $39
Skills	Standalone code transformations (error handling, tests, etc.)	Free - $49

Official Free Extensions

Get started with our free official extensions in official_extensions/:

Extension	Hook Point	Description
decomposition	before:design	Break complex features into subtasks
api-contract	before:implementation	Define API boundaries and contracts
code-review	after:implementation	Senior engineer review with feedback

Install them:

cp -r official_extensions/* ~/.coding-factory/extensions/agents/
acf extensions list

Quick Start: Create Your Own

1. Create extension directory

mkdir my-agent && cd my-agent

2. Create manifest.yaml

name: my-agent
version: 1.0.0
type: agent
author: Your Name
description: What it does
license: free
hook_point: "after:implementation"
agent_class: MyAgent

3. Create agent.py

from dataclasses import dataclass, field
from typing import Any

@dataclass
class AgentOutput:
    success: bool
    data: dict[str, Any]
    errors: list[str] | None = None
    artifacts: list[str] | None = None
    metadata: dict[str, Any] = field(default_factory=dict)
    agent_name: str | None = None

class MyAgent:
    def __init__(self, llm: Any, **kwargs):
        self.llm = llm
        self.name = "my-agent"

    def run(self, input_data) -> AgentOutput:
        context = input_data.context
        # Your logic here
        return AgentOutput(
            success=True,
            data={"result": "done"},
            agent_name=self.name
        )

4. Install and test

cp -r my-agent ~/.coding-factory/extensions/agents/
acf run "Build an API" --auto-approve

See full guide with profile and RAG kit examples →

Submit to the Marketplace

Share your extensions with the community and earn from your work.

1. Create an account

Sign up at https://agentcodefactory.com/signup.html

2. Verify your email

Check your inbox and click the verification link. This unlocks API access and marketplace submissions.

Didn't receive it? Log in and go to Settings to resend.

3. Create an API key

Go to Settings → Platform API Keys and click "Create Key". Copy it immediately—it's only shown once.

export ACF_MARKETPLACE_API_KEY=acf_sk_xxxx...

4. Build your extension

Create a directory with:

my-extension/
├── manifest.yaml    # Required: name, version, type, description
├── agent.py         # Your code
└── README.md        # Documentation

5. Submit for review

acf marketplace submit ./my-extension --price 15.00
# Use --price 0 for free extensions

Trust Badges

Genuine Official ACF Extension

Checked Security Reviewed

Trusted Proven Contributor

Revenue

• You keep 82.35% of each sale
• 17.65% covers payment processing (Stripe/PayPal) and infrastructure
• Payments via PayPal or Stripe

---

Architecture

~/.coding-factory/
├── extensions/          # Installed extensions
│   ├── agents/
│   ├── profiles/
│   ├── rag/
│   └── skills/
└── memory/              # Learning from past runs

Links

• Marketplace
• Developer Guide
• Documentation
• Main Platform

License

MIT License - Free for personal and commercial use.