The most-starred software repo in GitHub history is an open-source self-hosted AI agent that connects 50+ apps without external API calls. We've been running it for 3 months. Here's our honest take.
July 2, 2026 — OpenClaw crossed 210,000+ GitHub stars. That's more than React (230K), more than Vue (207K), more than TensorFlow (186K). It's now the most-starred software repository in GitHub history. And it's an open-source self-hosted AI agent that runs on your laptop.
We've been running OpenClaw in production for 3 months — for the Stashed Instagram pipeline, the Maruthi Jewellers WhatsApp agent, the 2BFT Academy enrollment, and 12 of the 100+ systems on /agentic-systems. Here's the honest take.
OpenClaw is a self-hosted AI agent framework. You download a single binary, run it on your machine or a small VPS, and it gives you an agent that can call 50+ tools (Gmail, Notion, WhatsApp, Google Calendar, Telegram, Slack, GitHub, Postgres, and more) without sending your data to anyone.
It has a built-in memory system, a small DSL for chaining actions, a web UI for inspecting runs, and an MCP server so Claude, GPT, or Gemini can use it as a tool. The whole thing is ~80MB. Runs on a $5/month VPS. No external API calls unless you explicitly add them.
Stashed Instagram pipeline: every Monday morning, OpenClaw pulls the week's product photos from a Google Drive folder, generates 6 captions in our brand voice using Claude, queues them in a Notion calendar, and posts them on schedule. Zero human in the loop unless we want to override. 30 minutes saved per week, 2BFT's most-used workflow.
Maruthi Jewellers WhatsApp agent: customer sends a photo of a gold piece, OpenClaw runs it through Nano Banana 2, identifies the design, checks inventory in our Google Sheet, and replies with price + availability. Handles ~40% of inbound WhatsApp without a human. 24/7.
2BFT Academy enrollment: when someone fills the Academy form, OpenClaw sends a welcome email, adds them to the Discord, schedules a 1:1 onboarding call if they want it, and tags them in our CRM. The full pipeline runs in 90 seconds.
Honestly: long-running stateful workflows. If you need a 30-step process that spans days, you want LangGraph or Mastra — those handle state better. For anything under 10 steps, OpenClaw is faster to set up and easier to debug.
It also doesn't do multi-agent orchestration well. If you need a researcher + writer + critic pipeline (which we do for some content), you'll want CrewAI or AutoGen. OpenClaw is best for single agents doing useful work.
We've tried 12 of them. The ones that work great: Gmail, Notion, Google Calendar, WhatsApp (via the Business API), Telegram, GitHub, Slack, Google Drive, Sheets, Postgres. The ones that work but are flaky: Twitter/X (rate limits), Discord (permission model is weird), Airtable (their API changed twice in 6 months). The ones we don't use: Stripe, Shopify, Salesforce (no Indian SMB need yet).
Yes, if: (1) you handle any data you don't want leaving your machine, (2) you're running a small business or team, (3) you don't have time to set up LangGraph, (4) you want to deploy in a day not a month. The cost is essentially free (compute only), the learning curve is two afternoons, and the docs are solid.
No, if: (1) you need 100+ step workflows with state, (2) you need production-grade observability, (3) you have a dedicated infra team that prefers to wire Kubernetes. Use LangGraph or Mastra for those.
The 12 of the 100+ systems on /agentic-systems that we ship with OpenClaw: WhatsApp auto-replier, Instagram DM qualifier, content cross-poster, daily morning briefing, email-to-task extractor, cart abandonment email, and 6 more. Each one is buildable in an afternoon.
If you're new to agents, start with OpenClaw + Claude Sonnet 5. You'll have a working WhatsApp agent in 2 hours.
2BFT is a 2-person studio in Vaniyambadi, Tamil Nadu. We test every major model on day one, ship real agentic systems, and teach builders across India. Free Academy, free newsletter, 200+ curated resources, 100+ copy-ready systems.
