marpy

marpy is a browser-based AI coding IDE and deployment platform built specifically for Python developers, created by Fiscus Technology LLC. It targets the pain points that generic AI coding tools miss — destructive database migrations, stale package versions, hardcoded secrets, and unpredictable deployments — by wrapping the AI assistant in opinionated guardrails before any damage is done.

What It Is

marpy combines a Monaco-based browser IDE, a managed MariaDB database layer, containerized deployments, and an AI coding assistant into a single Python-native platform. Unlike general-purpose AI coding tools that treat Python as a secondary runtime, marpy is built around Flask, FastAPI, and Django workflows from the ground up. The platform intercepts risky LLM-generated actions — such as dropping tables, pinning stale packages, or writing secrets to files — and either blocks them or rewrites them into safe, reviewable alternatives.

How the Guardrails Work

The core differentiator is a set of runtime intercepts that run between the AI assistant and the actual project environment:

• Migration safety: All schema changes are routed through versioned Alembic migrations. Destructive DDL (DROP TABLE, destructive ALTER) on production is blocked at the gate; ad-hoc DDL is rewritten as a reviewable migration file.
• Package freshness: Every pip install is intercepted, resolved to the current PyPI version, and the package's real documentation is indexed in the background so the assistant writes against the live API.
• Sandbox isolation: The project terminal is shell-wrapped to the project root, preventing path traversal outside the project tree.
• Secrets vault: Secrets are stored in a managed vault and injected as environment variables at runtime; the LLM never has read access to secret values in files.

Deployment Model

marpy uses date-based deployment tags (e.g., 202603061430) rather than semantic versioning, triggering containerized builds with a readable deploy log. The platform handles SSL, routing, and process health — described on the site as "like Vercel, but for your Python backend." Managed MariaDB instances include backups and point-in-time recovery, explicitly rejecting the container-local data pattern that causes data loss on restarts.

AI Integration and BYOK

The platform supports two AI usage modes. Users can bring their own API keys (BYOK) for Claude, GPT-4, Gemini, or any OpenAI-compatible endpoint; marpy routes requests via LiteLLM and never bills for tokens in this mode. Alternatively, users can rely on marpy-managed AI tokens included in each plan tier, with transparent overage billing. RAG context injection is included with the managed AI option.

Target Audience and Positioning

marpy explicitly positions itself against JavaScript-first platforms like v0, Replit, and Lovable, arguing that those tools bolt Python on as an afterthought. The site, written in first-person by a self-described "Python dev who is tired of TypeScript-first AI tools," frames the product as opinionated infrastructure for developers who want to ship real Python backends — not toy REPLs or serverless JS wrappers. The FAQ notes it is not aimed at developers whose entire backend is serverless functions behind a JS framework.

Current Status

marpy is in early access as of 2026, with a waitlist signup on the homepage. The site footer credits Fiscus Technology LLC and notes the platform itself is built with Flask and MariaDB. No public version number or changelog is surfaced, but the pricing page is live with defined tiers including a free Starter plan requiring no credit card.