Codonova 🤖

Autonomous Software Development System powered by deep AI agents,a Neo4j knowledge graph, ChromaDB memory, and a real-time React dashboard.

What It Does

Codonova autonomously:

1. Plans — Analyzes requirements, extracts features, decomposes into tasks with dependencies
2. Develops — Generates production-quality Python code using Gemini 2.5 Flash
3. Tests — Auto-generates and runs pytest test suites using Groq Llama 4 Scout
4. Debugs — Performs root-cause analysis and applies fixes using Gemini
5. Evaluates — Scores output 0-10 on correctness, quality, completeness
6. Self-corrects — Retries failing code up to 3 times with critique-augmented prompts
7. Learns — Stores successful patterns in ChromaDB for future context injection

Architecture

┌─────────────────────────────────────────────────────┐
│                  React Dashboard                     │
│  [Task Graph] [Agent Feed] [Code] [Scores] [Insights]│
└──────────────────────┬──────────────────────────────┘
                       │ REST + WebSocket
┌──────────────────────▼──────────────────────────────┐
│                  FastAPI Backend                     │
│                                                     │
│  PlanningAgent (Gemini)  → DecomposesRequirements  │
│  Scheduler               → Routes tasks             │
│  DeveloperAgent (Gemini) → Generates code           │
│  TestingAgent (Groq)     → Writes+runs tests        │
│  DebuggingAgent (Gemini) → Fixes failures           │
│  EvaluatorAgent (Groq)   → Scores output            │
│  CorrectionEngine        → Self-correction loop     │
└────────┬─────────────────────────┬──────────────────┘
         │ Neo4j Cypher            │ ChromaDB
┌────────▼──────────┐   ┌──────────▼──────────────────┐
│   Neo4j Graph DB  │   │  ChromaDB Vector Store       │
│  (Task knowledge) │   │  (sentence-transformers)      │
└───────────────────┘   └──────────────────────────────┘

Quick Start

Prerequisites

• Docker & Docker Compose
• API keys (all free):
  • Google AI Studio → Gemini 2.5 Flash
  • Groq Cloud → Llama 4 Scout
  • OpenRouter → Fallback (optional)

1. Configure Environment

cd autonomousdev
cp .env.example .env
# Edit .env and fill in your API keys and Neo4j password

.env required fields:

GEMINI_API_KEY=your_key_here
GROQ_API_KEY=your_key_here
NEO4J_PASSWORD=choose_a_password

2. Start with Docker Compose

docker-compose up -d

Services start in order: Neo4j → ChromaDB → Backend → Frontend

3. Verify Services

Service	URL	Purpose
Dashboard	http://localhost:3000	React UI
API Docs	http://localhost:8000/docs	Swagger UI
Neo4j	http://localhost:7474	Graph Browser
ChromaDB	http://localhost:8001	Vector DB

4. Run Without Docker (Development)

Backend:

cd autonomousdev/backend
pip install -r requirements.txt
uvicorn main:app --reload --port 8000

Frontend:

cd autonomousdev/frontend
npm install
npm run dev  # runs on http://localhost:3000

Usage

Via Dashboard

1. Open http://localhost:3000
2. Enter a requirement in the launcher (right panel)
3. Click 🚀 Start Autonomous Pipeline
4. Watch the Task Graph panel fill with nodes
5. Monitor live events in the Agent Feed panel
6. View generated code in the Code Viewer
7. Download the ZIP export when complete

Via API

Plan only:

curl -X POST http://localhost:8000/api/plan \
  -H "Content-Type: application/json" \
  -d '{"requirement": "Build a REST API for a task management app"}'

Full pipeline:

curl -X POST http://localhost:8000/api/start \
  -H "Content-Type: application/json" \
  -d '{"requirement": "Build a REST API for a student grade management system with endpoints to add students, record grades, and calculate GPA. Use FastAPI and SQLite."}'

Check status:

curl http://localhost:8000/api/status/{project_id}

Download export:

curl -O http://localhost:8000/api/export/{project_id}

LLM Configuration

Agent	Provider	Model	Why
Planning	Gemini	gemini-2.5-flash	Deep reasoning
Developer	Gemini	gemini-2.5-flash	Code generation
Debugger	Gemini	gemini-2.5-flash	Root-cause analysis
Testing	Groq	llama-4-scout-17b-16e-instruct	Speed + volume
Evaluator	Groq	llama-4-scout-17b-16e-instruct	Speed + volume
Fallback (all)	OpenRouter	meta-llama/llama-4-scout:free	Rate limit backup

Neo4j Graph Schema

Project ──HAS_FEATURE──► Feature ──HAS_TASK──► Task ──HAS_SUBTASK──► SubTask
                                                 │
                                          ◄DEPENDS_ON
                                                 │
                                       CodeModule ──PRODUCED_BY──► Task
                                          │
                                     TestResult ──VALIDATES──► CodeModule
                                          │
                            Bug ──FAILED_BY──► CodeModule
                             │
                           Fix ──RESOLVES──► Bug
                                          │
                          EvaluationResult ──EVALUATES──► CodeModule
                                          │
                         LearningNode ──PRODUCED_BY──► Task
                              │
                         Agent ──LEARNED_FROM──► LearningNode

File Structure

autonomousdev/
├── backend/
│   ├── agents/
│   │   ├── planning_agent.py      # Gemini — requirement decomposition
│   │   ├── developer_agent.py     # Gemini — code generation
│   │   ├── testing_agent.py       # Groq   — test gen + execution
│   │   ├── debugging_agent.py     # Gemini — root cause + fix
│   │   ├── evaluator_agent.py     # Groq   — quality scoring
│   │   ├── correction_engine.py   # Self-correction loop orchestrator
│   │   └── scheduler.py           # Task polling + routing
│   ├── graph/
│   │   └── neo4j_client.py        # All Cypher queries
│   ├── memory/
│   │   ├── memory_store.py        # ChromaDB + sentence-transformers
│   │   └── context_retriever.py   # Similar context finder
│   ├── llm_client.py              # Shared LLM client (NEVER bypass this)
│   ├── main.py                    # FastAPI app + all routes
│   └── requirements.txt
├── frontend/
│   └── src/
│       ├── App.jsx                # Main dashboard
│       └── components/
│           ├── TaskGraph.jsx      # Force graph visualization
│           ├── AgentFeed.jsx      # Live WebSocket feed
│           ├── CodeViewer.jsx     # Syntax-highlighted code
│           ├── EvalChart.jsx      # Score charts
│           └── InsightsPanel.jsx  # Neo4j analytics
├── docker-compose.yml
├── PHASE_1_SUMMARY.md ... PHASE_5_SUMMARY.md
└── .env.example

Generated Output

The ZIP export (GET /api/export/{project_id}) contains:

• src/ — all generated Python source files
• requirements.txt — auto-detected dependencies
• README.md — project overview generated by Gemini
• test_report.json — full pytest results
• EVALUATION_REPORT.md — per-module quality scores

---

Built with ❤️ by Codonova — Autonomous AI Software Development