marpy.io - AI coding platform built specifically for the Python stack
by•
Marpy is a web-based IDE and AI coding assistant built specifically for the Python stack. It helps you go from idea to deployed app without wrestling infra, glue code, or half-baked JS-focused tools. Get Python-native autocomplete, refactors, and AI-generated modules that actually understand Django, FastAPI, and real-world backends. Marpy lets you prototype, iterate, and ship production-grade Python apps faster, all from your browser.
Replies
who is the primary user here. a Python developer who already has a local setup and VS Code configured is probably not switching to a browser IDE regardless of how good the AI is. the person who might actually love this is someone who knows Python but hasn't gotten deep into tooling yet. are you building for the experienced dev or the intermediate one because those are pretty different products
@ansari_adin Good point. Definitely beginner/intermediate. I feel like experienced devs already have tooling in place, I know I do. With that said, I use it to hand projects off to other devs in a more consistent way.
I think for a beginner/intermediate user like myself, this would be a great tool. I usually switch between Windows for personal use and Mac for personal/work, so a browser-based, beginner-friendly Python tool would be invaluable. Keep up the great work!
@satwik_karnati Sweet. Thank you very much!!
The Python-first framing is the right bet here. Most AI coding tools optimize for the demo case (a Next.js app that looks good in a video), but Python-heavy workloads — Django migrations, async FastAPI services, pandas pipelines that touch prod data — have failure modes that JS-focused autocomplete just doesn't understand.
Curious how you're handling the "idea to deployed app" gap when the idea involves something stateful — a Celery beat task, a background worker that needs a DB connection. That's where I've seen vibe-coding tools fall apart: the generated code runs locally but assumes an environment that doesn't exist in prod. Does Marpy's context model pick up on infra assumptions early, or does that surface at deploy time?
@eran_shayshon It starts the project off with the infrastructure setup with Docker/k8s, so there is essentially no drift between dev/prod.
A Python-native browser IDE makes sense if it reduces the setup friction between idea, backend, and deployed app.
@zact Yep. That's a big reason why marpy is opinionated out of the gate. It helps fill the gaps in getting python scaffolded and getting it from working locally to running in production.
Really interesting congrats for the launch
Framework-specific AST awareness for Django and FastAPI models is what separates this from generic AI tools. Most don't understand ORM relationships or dependency injection patterns. We've wasted hours cleaning up AI suggestions that looked plausible but broke with SQLAlchemy migrations. Does marpy maintain cross-file context, like tracking model schemas and router dependencies across the full project?
@retain_dev Yep! marpy parses your project's Python with AST and builds a live, framework-aware map of it. Your SQLAlchemy/Django models (columns, primary keys, foreign keys, and relationships) and your FastAPI routers follows their Depends() chains.
That map is fed to the assistant on every request and to inline completions as you type, ranked to whatever file you're working in, so suggestions are grounded in your actual cross-file schema and dependency wiring instead of
guessed from a single open file. It refreshes incrementally as you edit, so it should stay accurate on bigger projects.
On the SQLAlchemy migration pain specifically: marpy statically analyzes every migration and hard-blocks destructive operations (DROP/TRUNCATE) against production. This is a deliberate platform guardrail, not an optional setting. So the "plausible but broke" class of suggestion is exactly what it's designed to catch.
Interesting angle focusing specifically on the Python stack instead of trying to be an “everything IDE”.
Feels like more AI tools are winning now by going deeper into a niche instead of broader.
@roki_014 Definitely, yes. And Python lends itself verrrry well to LLMs - so it's a good language to tackle.
The "understands Django and FastAPI ORM relationships" claim is the interesting one — does that come from reading my models.py statically, or does it need a live DB connection to know the actual schema? Generic AI tools fall over right here: they'll happily generate a query against a relation that doesn't exist. Curious how deep the model-awareness goes before it starts guessing.
@sounak_bhattacharya It essentially keeps your models.py in memory as a reference. I don't let it guess, if it can't find a concrete answer it will ask instead of guess or assume.