Starts Nov 23Live Cohort 3 of Vibe Coding & AI Agents (Nov 23, 18:30 MSK): 3 weeks, 9 live hands-on sessionsCohort 3 (Nov 23) →
ReleaseInsight · On RadarSimon Willison→ original

Author of mod_wsgi Uses AI to Write Muru, a New Python Application Server in Rust

Graham Dumpleton, the author of mod_wsgi, is building Muru, a Rust-based server for Python applications on WSGI and ASGI, written with the help of AI. It targets both development and production: according to his notes, it includes local HTTPS, canary releases, and scale-to-zero. In three weeks, the author has published 18 detailed posts about the project. Python developers should take a closer look, since this server draws on 20 years of experience.

AI-processed from Simon Willison; edited by Hamidun News
Author of mod_wsgi Uses AI to Write Muru, a New Python Application Server in Rust
Source: Simon Willison. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceSimon Willison ↗
Editor-in-Chief
Key metric #1Muru supports both WSGI and ASGI in a single server
Key metric #2Nearly one post per day, ranging from HTTPS in development to Kubernetes
Key metric #3The author says AI does much of the work, but he decides what is "not yet good enough"

A Python production server is something most teams choose once and then leave alone for years. That's why any new project in this area deserves attention, especially when it comes from someone who has worked on the subject for two decades. Graham Dumpleton created mod_wsgi 20 years ago and has been thinking about the best ways to run Python applications ever since. Now he is building Muru, a new server in Rust, written with heavy help from AI.

What Muru is and who it is for

Muru aims to run applications under two standards: WSGI (the classic synchronous interface behind Django and Flask in their traditional setup) and ASGI (asynchronous, for FastAPI and similar frameworks). According to the author, the goal is to provide the best set of tools for both local development and production. One of the notes is dedicated to how WSGI and ASGI work together within a single server.

The project's code is on GitHub under the muruhq organization. It is not a ready-made "gunicorn killer," and it is too early to talk about maturity: the project is very new, and the source is a set of links compiled by Simon Willison, not documentation or benchmarks. The material contains no performance figures, so there is nothing yet to compare Muru against uvicorn, gunicorn, or granian.

What the author's notes show

Dumpleton publishes a detailed development diary: between November 4 and 21, 18 posts appeared, almost one a day. Simon Willison notes that the author followed the same pattern with the wrapture project. The headlines give a sense of the problems he is tackling:

  • Worker internals. The posts "looking inside the workers" and "how it fits together" describe the internal architecture, while "import once, fork many" hints at a technique familiar from prefork servers: the application is imported once, and then the process is multiplied via fork.
  • Development. There are notes on HTTPS for local work and on a toolbar on every page, including request tracing.
  • Operations. The headlines promise releases without restarts, canary rollouts without a service mesh, integration with Kubernetes, autoscaling based on worker metrics, and scale-to-zero.

Implementation details are best read in the posts themselves. This article relies on their titles and Willison's summary, so specific capabilities should be verified against the project's documentation.

AI as a tool, not the decision-maker

Willison also recommends reading the post "AI as a lifestyle choice." In it, Dumpleton writes that AI can do a lot of the surface work, but he remains the one who decides what is still not good enough, and twenty years of mistakes with mod_wsgi are what make that judgment useful. He is prepared to keep polishing Muru until it becomes something he would trust himself and that others could trust too.

For developers, this is a telling example of AI-assisted development in systems software. It is not vibe coding for a demo: a production server demands careful handling of processes, networking, and failures, and the author takes on responsibility for the quality bar. The practical habit is also worth noting: a public development diary helps the author keep track of context, and helps future users understand why the server is built the way it is.

What to do with this

If you run Python services on Kubernetes and are tired of assembling separate tools around gunicorn or uvicorn for releases, canaries, and autoscaling, it makes sense to keep an eye on Muru. For now this is more an observation than a recommendation to adopt: the project is young, there are no independent tests, and the completeness of the stated features has not been verified. But the core idea, a single server that understands the release lifecycle and what is happening inside it, in both development and production, is interesting enough to look at the repository and the author's first two or three notes.

Story

Key Questions Answered

What Muru is and who it is for?

Muru aims to run applications under two standards: WSGI (the classic synchronous interface behind Django and Flask in their traditional setup) and ASGI (asynchronous, for FastAPI and similar frameworks). According to the author, the goal is to provide the best set of tools for both local development and production. One of the notes is dedicated to how WSGI and ASGI work together within a single server. The project's code is on GitHub under the muruhq organization.

What the author's notes show?

Dumpleton publishes a detailed development diary: between November 4 and 21, 18 posts appeared, almost one a day. Simon Willison notes that the author followed the same pattern with the wrapture project. The headlines give a sense of the problems he is tackling: Worker internals.

What are the key details on AI as a tool, not the decision-maker?

Willison also recommends reading the post "AI as a lifestyle choice." In it, Dumpleton writes that AI can do a lot of the surface work, but he remains the one who decides what is still not good enough, and twenty years of mistakes with mod_wsgi are what make that judgment useful. He is prepared to keep polishing Muru until it becomes something he would trust himself and that others could trust too. For developers, this is a telling example of AI-assisted development in systems software.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…