Hermes Agent crawl4ai plugin for self-hosted AI agent web extraction

[mblauser]

I built hermes-plugin-crawl4ai to wire crawl4ai into Hermes's web_extract tool as a plugin-registered backend. It calls crawl4ai's /md endpoint for clean markdown extraction, handles JS-rendered pages, and integrates with Hermes's multi-profile architecture.

The blocker: a core load-order bug in Hermes prevents plugin-registered web providers from being selected as backends, even when they're correctly installed and registered. Plugin discovery runs after backend selection, so the provider never surfaces. The symptom is a silent fallback to whatever non-plugin backend is configured (e.g. searxng) with a "search-only backend" error on extract calls.

I submitted a fix in PR #67309. It's a minimal 2-line change that moves plugin discovery before the shared backend-resolution chokepoints. It's been tested end-to-end against a live crawl4ai instance and includes a regression test for the cold-registry edge case. The PR has labels triaged and some initial activity, but it needs a project maintainer to do a final review and merge.

If you're a self-hosted Hermes + crawl4ai user (or know someone on the NousResearch team who maintains Hermes), could you help flag this PR to a maintainer? The fix is small, well-scoped, and unblocks a fully self-hosted search-and-extract stack that a lot of self-hosters want.

Repository: https://github.com/mblauser/hermes-plugin-crawl4ai
PR: NousResearch/hermes-agent#67309