12 GitHub Repos Every AI Marketer Should Bookmark

The pattern with open source for marketing teams: 80% of the value is in 12 repos, and most marketers haven't heard of any of them. Here's the shortlist I actually use or have shipped at clients.

Workflow + agents

• browser-use/browser-use — Python agent that drives a real browser. Beats Playwright for any task where a person would need to navigate, click, and read.
• Significant-Gravitas/AutoGPT — still the easiest way to chain multi-step research tasks for SEO and content briefs.
• stitionai/devika — open Devin alternative. Useful for scripted "agent does the boring part" workflows.

Prompt management + LLM orchestration

• langfuse/langfuse — prompt versioning, tracing, analytics. Self-host or cloud. Replaces $300/month vendor tools.
• PromptFoo/promptfoo — A/B test your prompts. Honest evals, not vibes.
• BerriAI/litellm — one API across 100+ LLM providers. Switch from GPT-4 to Claude to Gemini without rewriting code.

Scraping + SEO

• unclecode/crawl4ai — LLM-optimized web crawler. Pulls clean markdown ready for Claude or GPT input.
• mendableai/firecrawl — same idea, polished, has an API. Great for content briefs and competitor monitoring.
• apify/crawlee — battle-tested scraping toolkit if you need volume.

Content + analytics

• Helicone/helicone — open observability for LLM calls. Cost, latency, errors, per-user breakdown.
• PostHog/posthog — full product analytics, A/B testing, session replay. The "everything bagel" of analytics, self-hostable.
• umami-software/umami — privacy-friendly web analytics. GA4 without the surveillance vibe.

How to actually use this list

Don't install all twelve. Pick one from each row (workflow, prompt management, scraping, analytics), spend a Sunday wiring them together for one real workflow you do weekly, and see if the time saved justifies the maintenance cost. If yes, scale up. If no, move on.

A self-hosted stack only beats SaaS when you have the time to maintain it. Otherwise, paid tools win on operational margin.