Modverse #55: Mojo 1.0 Beta, Community Mojo Libraries, and Real-Time Patient Conversations Powered by MAX

Last month’s 26.3 release made Mojo 1.0 Beta official, added Wan 2.2 video generation to MAX, and launched mojolang.org as Mojo's new documentation home. The community kept pace: Decimo v0.10.0, the Mojo Scientific Library, Thistle, and mojo-kafka all hit the modular-community channel, Hippocratic AI is running real-time patient conversations with MAX inference, and AI agents built a working Mojo app with zero Python on the backend. Here's the roundup.

Community Innovations

Here's what the community built with MAX and Mojo last month:

• Decimo v0.10.0: Yuhao Zhu updated his arbitrary-precision decimal library to Mojo v1.0.0b1 with the project's biggest release yet. New in v0.10.0: a REPL mode for the CLI calculator (installable via Homebrew), optimized Decimal128 with Python decimal.Decimal-compatible semantics, a BigFloat type backed by MPFR for arbitrary-precision floating point, and a new Rational type. Benchmarks in the release put Decimal128 on par with Rust's rust_decimal in performance and ahead in ULP accuracy. Install with pixi add decimo or find the code on GitHub.

• Mojo Scientific Library (MSL) v0.1.0: The first stable release of MSL, a Mojo port of the GNU Scientific Library covering numerical integration, differentiation, interpolation, and statistics routines. Now available on the modular-community conda channel. Code is on GitHub.
• Thistle: native crypto library in Mojo: Thistle is a pure-Mojo cryptography library with ED25519 and X25519 support. Now on the modular-community channel. The repo is available here.
• mojo-kafka: Apache Kafka client for Mojo: dvirarad built a Kafka client for Mojo over librdkafka, with a Pythonic producer and consumer API. Connects Mojo workloads to Kafka without a Python intermediary. The project is on GitHub.
• Light of Baldr: Adam Kruger published a collection of Mojo 1.0 repos covering a web stack, a GPU Aho-Corasick multi-pattern matching kernel, and security tooling. Find it at github.com/lightofbaldr.

💡 Building something with MAX or Mojo? Share it in the Community Showcase and we may feature it here.

Modular Making Waves

• Modular 26.3: Mojo 1.0 Beta, video generation, and a new docs home. The 26.3 release on May 7 shipped the first official Mojo 1.0 beta (v1.0.0b1). It also added Wan 2.2 video generation to MAX, introduced TileTensor as the successor to LayoutTensor for GPU kernel development, launched mojolang.org as the dedicated Mojo documentation site, and brought in Distributed Tensor and safe closures.
• Hippocratic AI partners with Modular. Hippocratic AI is now running real-time patient conversations on MAX inference. Moving to MAX delivered 22% faster mean end-to-end latency on NVIDIA B300 hardware.

  

• Inkwell: why inference infrastructure matters as much as the model. Modular published a case study using Inkwell, its real-time AI storybook app, to show what the underlying inference platform makes possible. The argument is practical: model quality and inference quality are both constraints, and they're not independent. Check out the app at inkwell.modular.com.
• Translating code to Mojo with AI agents. A new blog post walks through the process of migrating Python and C++ code to Mojo using AI agent assistance, including where the workflow holds up and where human review still matters.
• How AI agents built a pure Mojo app and 10 libraries. A follow-up post covers mobin, a pastebin service built in Mojo with zero Python on the backend, written through AI agent collaboration. The post includes the 10 supporting libraries built the same way and a look at what Mojo 1.0 Beta development looks like from the AI-assisted side.

Open Source Contributions

If you've recently had your first PR merged, message Caroline Frasca in the forum (or @caroline_frasca in the Discord server) to claim your Modular swag! Check out the recently merged contributions from our community:

• Deftera186 [1]
• dylan-stark [1]
• gabrieldemarmiesse [1]
• gsmyridis [1] [2]
• kevenv [1]
• lightofbaldr [1] [2]
• mahendrarathore1742 [1]
• martinvuyk [1] [2] [3] [4]
• moseschmiedel [1]
• msaelices [1] [2]
• Olcmyk [1]
• piperchester [1] [2]

Modular News & Events: Stay Connected

• New contribution process for the Mojo stdlib. Starting with 26.3, non-trivial stdlib PRs require an issue first. Open one, wait for the accepted label, then write the code. The intent is to prevent contributors from putting real work into changes that stall or don't land, and the team commits to responding to issues within 48-72 business hours. Typos, doc fixes, and small bug fixes can still go straight to a PR. If you're looking for where to contribute toward Mojo 1.0, the post outlines three priority areas: applying stability markers across the stdlib, auditing custom-Self types on methods, and improving Python-Mojo interop. Full details in the forum announcement.
• Mojo 1.0 Beta tracking. Mojo 1.0.0b1 is out. Follow the nightly release threads and the Mojo roadmap to stay current as the stable release approaches.
• Monthly community meeting. The community meeting runs every month, virtual and open. Register on Luma and add yourself to the agenda if you’d like to present.