Skip to content

Jarvis2021/gantry

Repository files navigation

GantryFleet GantryFleet

The Headless Software Factory: From Voice to Production in 90 Seconds

License: Apache 2.0 Python 3.11+ FastAPI Status

"You don't write code. You direct the fleet."

GantryFleet is a local, voice-activated software factory. It transforms natural language (and optional design mockups) into deployed, tested applications—complete with GitHub Pull Requests for human review. It doesn't generate code snippets; it ships production systems.


Why GantryFleet?

Other AI Tools GantryFleet
Generate code snippets Deploy production apps
Require manual testing Self-healing CI/CD (3 retries)
Need copy-paste to run Live URL in 90 seconds
No audit trail Black Box evidence
Push directly to main PR workflow for oversight
Single-shot generation AI Architect Consultation
Text-only input Text + Design Image input

The GantryFleet Guarantee

Guarantee What It Means
"No Touch" Build Code runs in isolated Docker containers—never on your host
"Green Light" Deploy Only passing audits get deployed
"Black Box" Evidence Every mission has cryptographic audit trail
"Junior Dev" Model Never pushes to main—always opens a PR

The "Goldilocks" Architecture: Why We Chose Docker

To understand why GantryFleet is safer for Enterprise, compare how the current giants handle code execution:

Platform Approach Trade-off
Claude Remote Sandbox Safe, but disconnected from your local DB, private keys, network
Cursor Local Host (Bare Metal) Full access, but AI mistakes affect your actual machine
GantryFleet Local Containerization Best of both worlds

GantryFleet takes the third path: we build over Docker containers.

1. Security vs. "The Rogue Agent"

Unlike Cursor, which requires you to trust the AI with your file system, GantryFleet operates in a Zero-Trust environment:

  • Every task spins up a fresh, isolated Docker container
  • The Agent has "Root" access only inside that disposable box
  • If the AI makes a mistake or tries to delete a system file, it destroys a temporary container that didn't exist 5 minutes ago
  • Your laptop remains untouched

2. Integration vs. "The Cloud Wall"

Unlike Claude, which is stuck in the cloud, GantryFleet runs on your network:

  • Connect to your local Postgres database
  • Ping your internal staging servers
  • Access your private package registry
  • Enterprise Win: You get the safety of a sandbox without losing access to your internal tools

3. The "Works on My Machine" Guarantee

Enterprises love Docker because it enforces consistency. GantryFleet doesn't just hand you code; it hands you a verified Docker image. If GantryFleet says "The tests passed," it means they passed in a clean, standard environment—not just on a developer's messy laptop. This makes the handoff to CI/CD trivial.


Key Capabilities

1. AI Architect Consultation

GantryFleet doesn't guess. The AI Architect Agent analyzes your request, proposes a plan, asks clarifying questions, and breaks complex projects into iterations before building.

You: "Build me a LinkedIn-style professional network"

GantryFleet (AI Architect): "This is a 3-iteration project:
  • Iteration 1 (NOW): Core UI with mock profiles, feed layout
  • Iteration 2 (LATER): Authentication + real user data
  • Iteration 3 (LATER): Messaging + connections
  
  Shall I start with Iteration 1?"

You: "Yes, proceed"

[WebSocket] ARCHITECTING → BUILDING → DEPLOYING...

GantryFleet: "Live at https://linkedin-clone.vercel.app — PR opened."

2. Design Mockup Matching (95% Visual Fidelity)

Upload a screenshot, mockup, or sketch along with your request. GantryFleet's Vision-enabled Architect analyzes the image and replicates it with 95% visual accuracy:

  • Layout: Exact column/row structure and spacing
  • Colors: Precise hex values extracted from the image
  • Typography: Font sizes, weights, and line-heights
  • Components: Button styles, input fields, cards, shadows
  • Polish: Hover states, transitions, border-radius

The generated repo includes your mockup in the README as a "Design Reference."

3. Famous App Design Systems

Tell GantryFleet to "build something like LinkedIn" and it applies the exact design system—colors, fonts, layouts—from our FAMOUS_THEMES database.

Supported themes: LINKEDIN, TWITTER, INSTAGRAM, FACEBOOK, SLACK, SPOTIFY, NOTION, AIRBNB

4. 3-Tier Model Architecture (Never Fail on Simple Apps)

GantryFleet uses a multi-model fallback strategy to maximize success rate:

Tier Model Purpose
Tier 1 Claude 4 Sonnet Most capable, handles complex apps
Tier 2 Claude 4 Sonnet Balanced performance, reliable
Tier 3 Claude 3.5 Sonnet Battle-tested, production-proven

Result: Simple apps NEVER fail. Complex apps get 3 chances with progressively different models.

5. Zero-Trust Fabrication

  • Isolation: Every build happens in an ephemeral Docker Pod (512MB RAM, 180s timeout)
  • Self-Healing: If tests fail, the Architect analyzes logs and retries (up to 3x)
  • Policy Gate: Scans for secrets, forbidden patterns, and security issues

6. Real-Time WebSocket Updates

Subscribe to WS /gantry/ws/{mission_id} for live status: ARCHITECTING → BUILDING → DEPLOYING → DEPLOYED


Quick Start

Prerequisites

  • Docker Desktop
  • Python 3.11+
  • AWS Bedrock access (Claude 3.5 Sonnet)
  • Vercel & GitHub tokens

1. Clone and Configure

git clone https://github.com/YOUR_USERNAME/gantry.git
cd gantry
cp .env.example .env
# Edit .env with your credentials

2. Start the Fleet

# Start all services
docker-compose up -d

# Or run locally for development
pip install -r requirements.txt
uvicorn src.main_fastapi:app --host 0.0.0.0 --port 5050

3. Open the Console

open http://localhost:5050       # Web UI
open http://localhost:5050/docs  # OpenAPI docs

4. Build Something

# Via API
curl -X POST http://localhost:5050/gantry/voice \
  -H "Content-Type: application/json" \
  -d '{"message": "Build a todo list app with dark mode"}'

# Or use the Web UI at http://localhost:5050

Tech Stack

Layer Technology Why
API FastAPI (async) Non-blocking, WebSocket native
Auth Argon2 + TokenBucket Memory-hard hashing, per-user rate limiting
AI AWS Bedrock (Claude 3.5) AI Architect Agent
Execution Docker (via proxy) Zero-trust isolation
Deploy Vercel CLI Instant global CDN
Publish GitHub API PR workflow, never push to main
Storage PostgreSQL Missions, conversations, audit trail

Architecture Overview

GantryFleet Architecture

View as Mermaid Diagram
flowchart TB
    subgraph Input["Multi-Modal Input"]
        WebUI["Web UI (Chat + Image)"]
        Voice["Voice / Siri"]
        API["REST API"]
    end

    subgraph Core["GantryFleet Core"]
        Consultant["AI Architect Agent<br/>(Iteration Planning)"]
        Architect["AI Architect<br/>(Code Generation)"]
        Policy["Policy Gate"]
    end

    subgraph Foundry["Docker Foundry"]
        Pod["Ephemeral Pod"]
        Audit["Self-Healing Audit"]
    end

    subgraph Output["Production Output"]
        URL["Live URL (Vercel)"]
        PR["Pull Request (GitHub)"]
        Evidence["Audit Trail"]
    end

    Input --> Core --> Foundry --> Output
Loading

What's Built vs. What's Planned

✅ Production Ready

Watch the Demo: Download Demo Video

  • AI Architect Consultation with iteration planning
  • Self-healing builds (3 retry attempts)
  • Zero-trust Docker isolation
  • Vercel deployment with live URLs
  • GitHub PR workflow (never pushes to main)
  • Famous app clone themes (LinkedIn, Twitter, etc.)
  • Design image input (mockups included in repo)
  • WebSocket real-time updates
  • Argon2 authentication + TokenBucket rate limiting
  • Professional UI generation (Staff Engineer quality)

🔮 Roadmap (Future Iterations)

These features are in the research/planning phase. For early access or enterprise engagement, contact the author.

Feature Description Status
Multi-Agent Architecture Fractal agents that spawn sub-agents Subscription
Vector Memory Persistent embeddings (Supabase/Pinecone) Subscription
Temporal Orchestration Workflow engine for complex builds Subscription
arc-core-sdk Injection Generated apps include internal agents Subscription
Multi-Channel Slack, Discord, Telegram integration Subscription
OAuth/OIDC Enterprise SSO support Subscription

Enterprise & Advanced Engagement

GantryFleet is open-source under Apache 2.0—free for personal and commercial use.

For advanced features, custom integrations, or enterprise deployment:

📧 Contact: pramod.voola@gmail.com
👤 Author: Pramod Kumar Voola

Engagement Description
Subscription Priority support, custom themes, dedicated infrastructure
Enterprise On-premise deployment, SSO integration, SLA
Custom Development Multi-agent architecture, vector memory, custom skills

API Reference

Core Endpoints

Endpoint Method Description
/ GET Web UI
/health GET Health check
/docs GET OpenAPI documentation

Consultation Loop

Endpoint Method Description
/gantry/voice POST Start/continue consultation
/gantry/status/{id} GET Get mission status
/gantry/missions GET List all missions

WebSocket

Endpoint Description
WS /gantry/ws/{id} Real-time mission updates

Full API documentation at /docs when running locally.


Security

  • Docker Proxy: No direct socket access (tcp://docker-proxy:2375)
  • Pod Limits: 512MB RAM, 50% CPU, 180s timeout
  • Policy Gate: Scans for secrets, forbidden patterns
  • Audit Trail: Every build recorded in missions/

Contributing

See CONTRIBUTING.md for guidelines.


License

Licensed under Apache License 2.0—the same license used by Kubernetes, Terraform, and Swift.

  • ✅ Free for personal and commercial use
  • ✅ Safe for enterprise adoption

See LICENSE for details.


GantryFleet — The Headless Software Factory

Voice → AI → Docker → Deploy → PR

Built by Pramod Kumar Voola

About

GantryFleet: The Headless Software Factory. A CLI-first autonomous agent that turns voice/text commands into architected code, self-healing tests, and deployed Pull Requests. Stop chatting. Start shipping.

Topics

Resources

License

Contributing

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors