# Stackmaven

what it does

stackmaven is a curated catalog of dev and ai tools, models, and frameworks, with a news layer covering what just shipped, and a stack recommender on the way.

• catalog: every tool, model, and framework has a hub page: overview, pricing, alternatives, and an opinionated stackmaven_verdict paragraph. human-reviewed before publish, no ai-generated slop, no seo directory filler
• news: articles connect back to catalog hubs as spokes: launches, beat reports, comparisons, 30/60/90-day follow-up scorecards, stack snapshots. the news exists to update the catalog, not the other way around
• stack recommender (v1.0): describe a project, get a structured stack back with every item linked to its catalog entry. free tier rate-limited, pro tier unlimited + private stacks + iterative recommender. ships in phase 3
• stackmaven pro (later): paid tier for unlimited recommender runs, extended depth, deep-cuts newsletter, early scorecards, and catalog api access

why i built it

“i feel behind on dev and ai tooling” is the dominant emotional state for working devs in 2026. returning devs, senior devs busy with management, indie hackers picking stacks, hobbyists trying to figure out where to start: all asking the same question, and all getting served by the same shape of bad answer.

existing catalogs are seo-spam directories with no editorial bar. stackshare has stale data and no news layer. nobody owns the position of “high-quality, opinionated, browsable reference + active news + recommender that actually picks.”

stackmaven is the version of that publication i’d want to read myself, and one i’m in an unusual position to build, because “show devs what’s out there” is literally the dna i’ve been running for fifteen years on traversy media.

it’s also deliberately not a personal-brand project. the publication speaks as stackmaven, not as me. the voice is neutral-editorial, like a wire service. opinion is segregated into clearly labeled verdict paragraphs. the brand can grow independently: sellable, spinnable, not bound to the personal name.

how it’s built

a quick tour:

• astro 6 + tailwind v4 (via @tailwindcss/vite) + mdx, deployed on vercel. node 24 runtime, pnpm, zod 4 for content-collection schemas
• content lives in git: every catalog entry and article is an mdx file with validated frontmatter. no cms, no database in v1. the recommender (v1.0) reads a build-exported /public/catalog.json manifest instead of querying a db at runtime
• resend for newsletter: single audience with topics-based segmentation. resend topics handle the multi-newsletter case for free, no per-newsletter pricing tier needed
• anthropic claude opus 4.7 drives drafting (offline pipeline: vendor changelogs + release notes → review queue) and (v1.0) the recommender
• pagefind for search: static, in-browser, mit-licensed. same pattern memcrate.dev uses; same reasoning
• json-ld on every page type: software catalog entries get SoftwareApplication schema, articles get NewsArticle, the recommender will get WebApplication. structured data is the cheapest seo lever for a publication that has to compete with established directories

what’s next

• phase 2 (now): catalog depth from 15 → 50 → 250 entries, sustained news cadence, 1K newsletter subs. daily tool-update workflow against vendor changelogs is the load-bearing automation here
• phase 3: recommender mvp, first sponsorship, affiliate revenue from tool partnerships (vercel, supabase, stripe etc.)
• phase 4: pro tier live, scale to revenue. v1.0 ships when the recommender lands; “v1” doesn’t equal “site shipped,” it equals “the recommender works”

nobody owns “the high-quality opinionated tooling reference + news + recommender” lane yet. that’s the bet: that the position is real, the editorial bar is the moat, and the recommender turns the catalog from a thing you browse into a thing you ask.