Guillermo Rauch's open source grants move toward AI agents

Guillermo Rauch handed out 35 $1,000 grants on September 7 in the v2 round. Not one of them funds a JavaScript framework, the category that used to define his grant list.

The last round, in November 2025, gave out 22 grants for general JavaScript tooling. This one grew to 35 grants, 13 more than last time, and every one of them targets agent infrastructure instead.

A thousand dollars each doesn't fund a runway. What it buys is a public list of 35 projects that one infrastructure funder picked out as the gaps worth filling right now.

The theme list doubles as a map of gaps in agent tooling. Four categories carry the label outright: "Postgres tooling for agents," "fast and accurate file search SDK for agents." Two more read "CLI for structural code exploration that agents love" and "MCPs for agents to automate popular creative apps." A fifth line funds three separate local-inference engines, including a pure-C MoE engine and one built for running large MoE models on Macs. Both are aimed at running big models without calling a hosted API.

Recipients include Kovid Goyal, Herrington Darkholme (ast-grep), Emil Kowalski, and Lars Kappert (knip). The list doesn't say whether any of the three local-inference grants have shipped code yet.