open-aidlc

OpenAIDLC: An Open-Source, Cost-Efficient, and Extensible Agentic Software Development Life Cycle Framework

---

🚀 What is OpenAIDLC?

OpenAIDLC is an open-source, declarative process framework and developer tooling suite designed to operationalize autonomous AI agents across the entire Software Development Life Cycle (SDLC) at least cost (<$0.50 per task).

While the software engineering industry is rushing to adopt AI coding agents, mainstream adoption by startups, agencies, and engineering teams faces three major bottlenecks:

1. The Token Cost Trap ($10–$25+ per Task): Monolithic agents pipe entire multi-turn bash and test trajectories into expensive frontier models (Claude 3.5 Sonnet / GPT-4o), making autonomous development cost-prohibitive.
2. Setup Friction & Integration Overhead: Configuring agent frameworks across disparate developer tools requires hundreds of hours of manual JSON-RPC plumbing, custom sandboxes, and permission scripting.
3. Multi-Software Tool Fragmentation & Artefact Drift: Modern teams rely on heterogeneous tools (Linear, Jira, Notion, Slack, Docker, Playwright, Appium) but lack unified lifecycle orchestration, causing code changes to drift away from requirements and deployment manifests.

OpenAIDLC solves these challenges through Cost-Aware Tiered Model Routing, an open Model Context Protocol (MCP) tool bus, autonomous repository onboarding (aidlc init --ai), and strict Git worktree isolation.

---

🏛️ System Architecture

OpenAIDLC is structured into a modular 6-layer engine:

flowchart TD
    subgraph L6["Layer 6: Autonomous AI Setup Agent (aidlc init --ai)"]
        Repo["Target Codebase"] --> FP["Fingerprinting Engine<br/>(Languages, Frameworks, Tests, CI)"]
        FP --> Int["Interactive 3-Question Alignment"]
        Int --> Scaff["Auto-Scaffold Config & Harnesses<br/>(aidlc.config.yaml, .mcp.json, AGENTS.md, CLAUDE.md)"]
    end

    subgraph L1["Layer 1: Cost-Aware Tiered Model Router"]
        Task["Development Task"] --> Router{"Cognitive Complexity<br/>Classifier"}
        Router -->|"Tier 0: $0.00"| SLM["Free Local SLMs (Ollama / Tree-sitter)<br/>AST Slicing, Linting, Syntax Parse"]
        Router -->|"Tier 1: <$0.05"| Flash["Sub-Cent High-Speed Models<br/>(Gemini Flash / Claude Haiku)<br/>Triage & Log Compaction"]
        Router -->|"Tier 2: <$0.40"| Frontier["Frontier Models (Sonnet / GPT-4o)<br/>Architecture & Multi-File Synthesis"]
    end

    subgraph L3["Layer 3: Multi-Tier Lifecycle & Worktree Isolation"]
        L1_Work["L1: Support Triage<br/>(Linear / Jira / Slack)"]
        L2_Work["L2: Dev-Fix Engine<br/>(Isolated Git Worktrees + CRG)"]
        L3_Work["L3: Architecture & RAG<br/>(Notion / Confluence Specs)"]
        QA_Work["QA: End-to-End Test Agent<br/>(Playwright Web + Appium Mobile)"]
    end

    subgraph L4["Layer 4: Policy-as-Code Governance"]
        Sync["DevOps Artefact Consistency Engine<br/>(Bidirectional Sync: Specs ↔ Code ↔ CI)"]
        Gate["Human-in-the-Loop Solution Approval Gate"]
        Cap["Strict Budget Guardrails (max_cost_usd: $0.50)"]
    end

    L6 --> L1 --> L3 --> L4

---

💡 Key Features

1. 💰 Cost-Aware Tiered Model Routing (97.2% Cost Reduction)

Instead of piping all prompts into expensive frontier LLMs, OpenAIDLC dynamically partitions tasks across three cognitive tiers:

• Tier 0 ($0.00 / Free Local SLMs): Local models (DeepSeek-Coder 7B via Ollama/vLLM) and Tree-sitter parsers handle AST slicing, formatting, and lint verification locally with zero API charges.
• Tier 1 (Sub-Cent Fast Models): High-throughput models (Gemini 1.5 Flash, Claude 3.5 Haiku) perform issue triage, log compaction, and PR categorization.
• Tier 2 (Frontier LLMs): Top-tier models (Claude 3.5 Sonnet, GPT-4o) are called only for high-level architectural design and complex multi-file patch generation.

2. ⚡ Autonomous AI Project Initializer (aidlc init --ai)

Eliminates setup friction by autonomously inspecting any existing Git repository, fingerprinting its build tools, test suites, and CI pipelines, conducting a brief interactive alignment interview, and scaffolding turnkey MCP toolchains in minutes.

3. 🔌 Universal Tool Bus via Model Context Protocol (MCP)

Provides native, zero-glue-code integration across the complete modern software stack:

• Issue Trackers: Linear, GitHub Issues, Jira
• Knowledge Bases: Notion, Confluence, Markdown PRDs
• Communication: Slack, Discord, Microsoft Teams
• Quality Assurance: Playwright (Web E2E), Appium (Mobile), Jest, PyTest
• Environments: Docker sandboxes, isolated Git worktrees

4. 🛡️ Safe Execution with Git Worktrees & Policy Governance

OpenAIDLC never commits directly to your working branch. All code modifications execute inside isolated Git worktrees (git worktree add). Configurable human approval gates ensure you review solution designs before code modifications take place.

---

📊 Empirical Cost Comparison: OpenAIDLC vs. Traditional Agents

Lifecycle Phase	Unoptimized Agents (OpenHands / Devin)	OpenAIDLC Tiered Architecture	Active Tier	Traditional Cost	OpenAIDLC Cost	Savings
1. Ticket Triage & Context Retrieval	Claude 3.5 Sonnet (250k input tokens)	Gemini 1.5 Flash (Compacted)	Tier 1 (Flash)	$0.75	$0.02	97.3%
2. Codebase AST Indexing & Slicing	Claude 3.5 Sonnet (1.2M raw tokens)	DeepSeek-Coder 7B via Ollama	Tier 0 (Local)	$3.60	$0.00	100%
3. Architecture & Solution Design	Claude 3.5 Sonnet (400k tokens)	Claude 3.5 Sonnet (Sparsified)	Tier 2 (Sonnet)	$1.20	$0.35	70.8%
4. Implementation in Worktree	Claude 3.5 Sonnet (2.5M multi-turn)	DeepSeek 7B + Claude 3.5 Haiku	Tier 0 + Tier 1	$7.50	$0.05	99.3%
5. Linting, Unit Testing & Fixes	Claude 3.5 Sonnet (800k tokens)	Local linter + DeepSeek 7B	Tier 0 (Local)	$2.40	$0.00	100%
6. Artefact Sync & PR Documentation	Claude 3.5 Sonnet (130k tokens)	Gemini 1.5 Flash	Tier 1 (Flash)	$0.40	$0.02	95.0%
TOTAL PER DEVELOPMENT TASK	4.28M tokens on Frontier LLMs	Compacted Tiered Routing	Multi-Tier	$15.85	$0.44	97.2%

---

🥊 Competitive Landscape: Why OpenAIDLC Stands Out

Dimension	OpenHands (OpenDevin)	SWE-agent (Princeton)	Aider	MetaGPT	OpenAIDLC (Ours)
Primary Focus	Docker sandbox dev	SWE-bench benchmark	Terminal pair coding	Roleplay simulation	Full-Lifecycle SDLC
Avg. Cost / Task	$10–$25+	$8–$15	$1–$5	$12–$30+	<$0.50 (Guaranteed)
Outer SDLC (Jira/QA)	❌ No	❌ No	❌ No	❌ No	✅ Full Native (MCP)
Autonomous Setup	❌ Manual Docker	❌ Python harness	❌ Manual Git	❌ Manual SOPs	✅ Turnkey (init --ai)
Execution Safety	Docker sandbox	Read-only tools	Direct Git commits	None	✅ Worktrees + CRG
Target Audience	AI researchers	Benchmark evals	Individual coders	Chat demos	Startups & Engineering Teams

---

📂 Repository Structure

open-aidlc/
├── config/                              # Declarative configuration blueprints
│   ├── aidlc_config_example.yaml        # Comprehensive production blueprint
│   └── templates/                       # Ready-to-use configuration templates
│       ├── lean-startup.yaml            # Budget-capped ($0.50), Linear + GitHub + Slack
│       └── enterprise-strict.yaml       # Strict governance, Jira + Confluence + Docker
│
├── docs/                                # Technical specifications & architecture
│   ├── ARCHITECTURE.md                  # Detailed 6-layer engine & DAG workflows
│   ├── TIERED_ROUTING.md                # Mathematical model routing & cost optimization
│   ├── MCP_SPECIFICATION.md             # Model Context Protocol integration standards
│   ├── ROADMAP.md                       # Open-source release milestones (Alpha -> GA)
│   ├── VALUE_PROPOSITION_AND_BENEFICIARIES.md # ROI, in-house build vs. buy, persona benefits
│   └── benchmarks/
│       └── COMPETITIVE_ANALYSIS.md      # Industry benchmark (OpenHands, SWE-agent, Aider)
│
├── examples/                            # Real-world usage workflows
│   ├── linear-docker-workflow.yaml      # Linear bug fix to Docker container test & PR
│   └── jira-worktree-workflow.yaml      # Jira ticket to isolated Git worktree patch
│
├── CONTRIBUTING.md                      # Guide for open-source contributors
├── CODE_OF_CONDUCT.md                   # Contributor Covenant standard
├── LICENSE                              # Apache 2.0 License
└── README.md

---

🛠️ Quickstart

1. View Configuration Blueprint

Inspect the sample configuration to see how OpenAIDLC defines model routing, spending caps, and MCP servers:

cat config/aidlc_config_example.yaml

2. Choose a Starter Template

Copy a template that fits your team's workflow:

# For high-velocity startups:
cp config/templates/lean-startup.yaml aidlc.config.yaml

# For SOC2 / regulated enterprise environments:
cp config/templates/enterprise-strict.yaml aidlc.config.yaml

---

🗺️ Project Roadmap

• v0.1 Alpha (Current): Declarative configuration schema, 6-layer engine specification, MCP tool bus, reference templates.
• v0.2 Beta: Core CLI runner (aidlc run), local Ollama AST slicer, isolated Git worktree management, and budget caps.
• v0.3 Preview: Autonomous onboarding wizard (aidlc init --ai), interactive terminal TUI, and OpenTelemetry tracing.
• v1.0 GA: SWE-bench Lite validation, turnkey GitHub Actions / GitLab CI integrations, and enterprise policy packs.

See the full Project Roadmap for details.

---

🤝 Contributing

We welcome contributions of all kinds! Whether you want to add new MCP connectors, improve model routing heuristics, or enhance documentation, please check out our Contributing Guide and Code of Conduct.

---

📜 License

OpenAIDLC is open-source software licensed under the Apache License 2.0.