Why Tako (vs Axum) ?

[ririsoft]

Hello,

I discovered Tako through a Rust Weekly newsletter and your post about it.

I am currently using Axum and reading your blog post I had hard time identifying major differences with Axum. It looks pretty much similar to me, and shares a lot of the design choices. Maybe it is a little bit simpler, lighter, easier to extend, which is a good thing to me, I like the minimal approach.

I am curious and wanted to understand your motivations and what made you unhappy with Axum and why you started Tako, if you mind sharing ?

(Sorry, not a bug nor a feature request, you might move it to GitHub discussions if you like).

[dancixx]

@ririsoft
Thanks for checking out the project and for the thoughtful question!

Tako started as a personal learning project—an excuse to dig into async internals and performance work. I like Axum a lot and still think it’s an excellent default in Rust. Tako isn’t trying to outdo it; it’s a small, minimal place to experiment and keep the core easy to reason about. More context here: Tako — a lightweight async web framework on Tokio and Hyper.

Why it exists (and for whom)
Tako aims to be a lightweight base for folks who enjoy poking at the internals, trying low‑level ideas, and trading a bit of ecosystem breadth for a codebase that’s easy to read and extend. If you value maturity, stability, and ecosystem integrations, Axum is great. If you want something lean for experimentation, Tako might be interesting.

What’s different right now

• No Tower dependency. Tako ships with a small built‑in plugin system so behavior can be extended directly.
• Minimal core, few deps. The goal is to keep things straightforward to follow and hack on.
• Perf‑oriented bits out of the box:
  • Integrated SIMD‑accelerated JSON.
  • Default tikv-jemallocator, which in simple local tests has looked better on allocation‑heavy, JSON‑centric endpoints.
• Familiar by design. Many choices feel similar to Axum on purpose—so you don’t have to relearn everything to try Tako—while keeping room for lower‑level experiments.

Where this might go next

• Quasi‑stable v1. Tighten the public API, document it clearly, and minimize churn after that.
• Targeted low‑level optimizations. Allocator tuning, cache‑aware paths, and—where it pays off—small hand‑tuned kernels (even tiny asm snippets) for specific hot spots.
• I/O exploration. Take a serious look at io_uring, mio, or even a lighter alternative to parts of Hyper if (and only if) they bring real‑world wins—while keeping the public API intact.
• Plugin ecosystem. Iterate on the plugin API and docs as real use cases show up, and make it easy to share community plugins.

Tako is still young. The intent is to move quickly while validating ideas, then stabilize the surface that proves useful. If Axum already does everything you need, that’s an awesome choice. If you’d like a lean codebase to tinker with and explore performance angles, I’d love to hear what you try.

I’m very open to feedback, benchmarks you care about, and suggestions (or small PRs). Thanks again for taking the time to look through the project!

[ririsoft]

@dancixx thanks for sharing !

It looks like a good plan and I hope you will have the fun and learnings you are looking for.

I have been using different http frameworks in various languages and I am used to switching from one to another so I will definitely give Tako a run. As an Axum user the transition looks very easy.

The minimal core, few deps and plugin ecosystem is definitely what I like in "micro http frameworks". Abandoning Tower is not a big deal at all for me (it is even a plus), since I never really jumped in: you have to learn the http framework + the networking framework (it reminds me of the age of Python Twisted).

I will be interested in your io_uring explorations: Since it is out I never had the chance to use it in "production" for my hobby projects due to lack of support in the ecosystem for the moment. However my usage is limited to hobby and family so I won't be able to provide valuable production feedback on that part.

Regarding the plugins here are my areas of interest (and might contribute to, no promise only working as a hobby here):

• Client IP address extractor (when used behind a proxy)
• HTTP Range requests (File server).
• OpenID Connect authentication
• Light and secure session management.
• Logging and monitoring : the Tracing and Axum (and Rust) experience has always been disappointing for me in this area.

So what is next for me ? Try Tako and report back my experience and findings :)

All the best to you.