Clojurists Together – Q2 2026 Open Source Funding Announcement

Source: Hacker News

$9 K USD Projects

• Ambrose Bonnaire‑Sergeant – Malli
• Dragan Djuric – Uncomplicate AI: Clojure LLM
• Cvetomir Dimov – SciCloj Documentation and Plotting Libraries

$2 K USD Projects

• Ingy döt Net – Gloat
• Shantanu Kumar – PluMCP

Ambrose Bonnaire‑Sergeant – Malli

In my previous Clojurists Together work for Malli, I improved the performance of validating recursive refs, bounding the amount of memory required for validation regardless of the depth of input values. This is implemented by eagerly expanding recursive schemas until recursion points are discovered, instead of lazily realizing and caching an unbounded number of new recursion levels as inputs require.

While this increases the reliability of long‑running systems by preventing memory leaks caused by validating large inputs, it introduced a drawback: more memory is required upfront during validator compilation. Metabase has been navigating this obstacle while testing the optimization. Although they are excited that validation of nested structures now uses constant memory, the upfront memory cost was uncomfortably high.

There are two main ways to tackle this problem:

1. Discover recursion points lazily – reduces initial memory use, but the maximum memory usage still grows over time as large inputs are validated, making JVM memory‑reservation harder to predict.
2. Reduce the maximum memory usage – the current optimization interfered with a custom Metabase optimization that shares schemas across usages. The root cause is that identical schemas were using distinct validators each time they were used. The solution is to ensure that references to the same Schema object and validator point to the same instance. This issue can be severe in the general case (likely why Metabase needed to patch it).

Project goal: Investigate ways to improve this situation so that systems like Metabase can benefit from constant‑memory recursive validators without incurring prohibitive upfront memory costs, ideally eliminating the need for custom work‑arounds that address Malli’s high memory usage.

Dragan Djuric – Uncomplicate AI: Clojure LLM

My goal is to provide a high‑performance local LLM (large language model) AI solution that supports mainstream open models freely available on Hugging Face and elsewhere. Think of something like llama.cpp, but (hopefully!) simpler and faster, with both GPU and CPU support baked in from the start.

Clojure LLM will deliver a high‑performance local LLM library with a simple, straightforward high‑level API and a “batteries‑included” implementation based on Clojure ONNX Runtime.

During this funding cycle I plan to deliver a first version that supports Google’s Gemma 3 model family and Google’s SentencePiece tokenizer. Gemma 3 is multimodal, offers excellent overall performance, and supports 140 languages out of the box. The implementation will tightly integrate with Clojure’s ecosystem (ClojureCUDA, Neanderthal, Deep Diamond, etc.). Gemma 3 comes in several sizes—from 270 M (text‑only) to 27 B parameters—allowing users to pick the model that fits their hardware.

Cvetomir Dimov – SciCloj Documentation and Plotting Libraries

SciCloj is a Clojure group whose goal is to help Clojure grow into new and uncommon domains, use cases, and professions. On the technical side, SciCloj develops tools and libraries geared toward data analysis, AI, documentation workflows, scientific computing, interop, and beginner‑friendly experiences. On the community side, it runs multiple work groups and weekly meetings (recently focused on tooling and AI), mentors open‑source contributors, and organizes conferences on specialized topics. These steady efforts culminated in the first online Clojure conference for data‑analysis stories, SciNoj.

SciNoj demonstrated the wide range of applications that Clojure already supports. This project aims to extend and improve the libraries in Noj, the entry point for data analysis in Clojure, and their documentation in three ways:

1. Extend plotting capabilities – add more back‑ends and support additional plot types.
2. Create a new library for interactive dashboards.
3. Expand the Noj book – provide more complete coverage of all Noj libraries and enforce a consistent chapter structure.

Ingy döt Net – Gloat

Gloat already compiles Clojure to:

• Go code (with related Go build files)
• Native binaries (cross‑compiled to ~25 targets)
• Wasm modules (browser + server)
• Shared libraries (with FFI bindings for 20+ languages)

Gloat aims to be a complete replacement/alternative to GraalVM’s native-image for Clojure, offering:

• Much faster build times
• Far broader platform reach
• 100 % Open‑Source licensing

The project works, but it still needs:

• Smaller, faster binaries
• More real‑world Clojure testing

Glojure is new and still rough around the edges, so the Gloat project uses the gloathub/glojure fork to advance both Gloat and Glojure in parallel, with strong upstream communication and a plan to push all improvements upstream.

Funding goals for Gloat/Glojure

1. Make binaries smaller and faster

   • Progress already made: a “tree‑shaking” mode for clojure.core that omits unused parts.

2. Pass more of the Clojure Compatibility Test Suite

3. Create tutorial documentation

   • How to use Gloat to integrate Clojure into Go projects
   • How to use Gloat instead of GraalVM to (cross‑)compile Clojure programs to native binaries, shared libraries, and Wasm modules

Shantanu Kumar – PluMCP

PluMCP (Model Context Protocol) is a pivotal component of the Agentic‑AI ecosystem. It is currently the only Clojure + ClojureScript project that implements both MCP client and server features, supporting STDIO and streamable HTTP transports with OAuth 2.1 integration.

• Used by the open‑source project eca.dev
• Implements MCP spec version 2025‑06‑18

Funding objectives

Implement the next spec version 2025‑11‑25 and improve documentation. Planned major changes:

• OpenID Connect Discovery 1.0 support for authorization‑server discovery
• Ability for servers to expose icons as additional metadata for tools, resources, resource templates, and prompts
• Enhanced authorization flows with incremental scope consent via WWW-Authenticate
• Validation of tool names according to the new spec
• Update `