Muru, Graham Dumpleton's Rust Server for Python Apps, Has 185 Open Issues and No Benchmarks
Software / news
Muru, Graham Dumpleton's Rust Server for Python Apps, Has 185 Open Issues and No Benchmarks
The mod_wsgi author's alpha project takes contributions as requirements rather than code, and its first post shows a 0.2 ms log line.

Graham Dumpleton, who started mod_wsgi about 20 years ago, is building Muru, an AI-assisted Rust server for Python web apps, and its GitHub repository carries 185 open issues against 68 stars and a self-declared alpha label.
Simon Willison linked the project on Nov. 21, calling attention to a run of development posts Dumpleton has published since Nov. 4. The repository is licensed Apache-2.0 and showed 742 commits on its develop branch, 0 forks and 0 watchers when we read it on Nov. 22.

What Muru does, according to its README
The README describes a "Routing edge, control plane and Python runtime for WSGI and ASGI applications." A Rust edge, in the same language as the toolkits in Raph Levien's Rust GUI survey, accepts client connections and routes them to Python worker processes. A Rust controller manages those workers as releases.
The README says "A new release, possibly on a different Python version, starts beside the old one, and traffic moves when it is ready." It lists blue/green deploys, gated canaries, previews and instant rollback, plus memory limits, soft and hard timeouts and support for free-threaded Python. It also lists an MCP server so coding agents can deploy, test and roll back, which puts it beside OpenSRE in the agent-operated infrastructure pile.
The install path is two commands: uv tool install muructl, then muructl run app:application, which serves on http://127.0.0.1:8080.
What the first post confirms, and what it defers
Dumpleton's first post, dated Nov. 5, shows plain WSGI apps, Flask and Django running without Muru-specific changes. A sample access log records 0.2 ms total, 0.1 ms in the app and 0.0 ms queued for an 18-byte response, from a setup of 1 worker and 4 threads.
The same post lists limits. It is alpha with pre-releases only, runs on Linux and macOS, needs CPython 3.11 or later, and its development mode listens on loopback and is "not meant to be exposed to the outside world." The daemon and container modes were not yet covered, and the import-time fork-safety check was deferred to a later post.
| Claim | Where it appears | What the sources show |
|---|---|---|
| Zero-downtime releases | README, Nov. 22 | Reload shown in the Nov. 5 post |
| Canaries, scaling to zero | README and later posts | Titles seen, posts not read by us |
| Production readiness | Nowhere | README says alpha |
| Benchmarks | Nowhere | None published in what we read |
The last row is the gap. A tool pitched at production hosting has published a 0.2 ms log line and no throughput or latency comparison with gunicorn, uvicorn or mod_wsgi, and Dumpleton's first post does not name those tools.
Contributions arrive as requirements
The README says Muru "takes contributions as requirements, not as code." Someone with a problem opens an issue describing the behaviour they want, and the maintainer implements accepted work. The repository page adds that only collaborators can create pull requests.
That policy explains part of the 185 open issues, since each is a request rather than a patch, though the page does not break the count down. Dumpleton's post "AI as a lifestyle choice" says the same thing from the other side: "So don't send me a patch, just describe the problem."
How the AI work is described
The README says Muru is "developed with extensive use of AI coding agents." Dumpleton's post on the subject gives no figure for how much code the AI wrote. He wrote that "deciding what isn't good enough yet is still down to me" and that he is "prepared to put in the hard work to keep polishing Muru until it is something I would trust."
He says he has programmed for more than 40 years and has worked on Muru about a month, seven days a week. He moved from a $20 to a $200 monthly plan before being accepted into the Claude for Open Source Program, which removed the cost concern. He wrote that "If Muru has problems, they are my problems."
The next marker is the 0.1.0 line on PyPI, where the README says commands and configuration may change between pre-releases.
Sources
More in Software
- 01Fast-Anything Reports a 116x Python Speedup, and 8 of Its 9 Benchmark Rows Run One JobThe MIT-licensed tool ports hot functions to Rust with a coding agent, and its README gives no run counts or variance.
- 02ClawDOS Runs a Claude Coding Agent on a 4.77 MHz IBM PC XT, With TLS Handed to a TunnelThe GPLv3 project keeps a 176 KB harness on the 8088 and offloads encryption to a second machine, leaving a launched program 200K of the 640K.
- 03Levien's Next-100 Rust GUIs Post Counts About 20 New Toolkits and Bets Against egui's Immediate ModeThe Linebender author credits AI rewrites for the surge; egui's own README concedes the trade-off he is pressing on.
- 04C++26 Is Reopened: The 13–9–12 Ballot, the Enforced Label That Answers It, and What Is Still OpenSutter's Búzios report says the committee kept contracts, added one paper, and will send C++26 to a second ballot with no published date.