✨ This week in AutoBot: ✨ docs(process): name the selector — an empty resu. Shipped 3 highlights, +2 contributors. All open source → https://github.com/mrveiss/AutoBot-AI/commits/Dev_new_gui?since=2026-09-02T09:00:03.220327Z

Don't just chat with one AI. Build a workforce. Compose multi-step agents on a visual canvas. Add modules: an agent-company (AutoBot LLC), a transcriber, codebase analytics (early). All open source, all on infrastructure you own. Your data. Your AI. github.com/mrveiss/AutoBot-AI

✨ This week in AutoBot: ✨ fix(automation): count landed files, not just su. Shipped 3 highlights, +2 contributors. All open source → https://github.com/mrveiss/AutoBot-AI/commits/Dev_new_gui?since=2026-08-26T09:00:04.342434Z

New post: The half of your Claude setup that can't be shared — and why it still needs to be a plugin Why your team's conventions need their own plugin marketplace, and the one-line… https://dev.to/mrveiss/the-half-of-your-claude-setup-that-cant-be-shared-and-why-it-still-needs-to-be-a-plugin-20dg

An agent that acts is only safe if you stay in the loop. AutoBot shows its plan and waits for your OK before it clicks, sends, or changes anything. Full autonomy when you want it — a hand on the wheel when you don't. Open source. Your data. Your AI. github.com/mrveiss/AutoBot-AI

✨ This week in AutoBot: ✨ fix(setup-wizard): bound the provision-fleet 409. Shipped 3 highlights, +2 contributors. All open source → https://github.com/mrveiss/AutoBot-AI/commits/Dev_new_gui?since=2026-08-19T09:00:03.226436Z

New post: Agents don't fail at writing code — they fail at everything around it Six months of running Claude Code agents on a large codebase: the four process failures that actually cost me, and the ski… https://dev.to/mrveiss/agents-dont-fail-at-writing-code-they-fail-at-everything-around-it-46ah

Manuals are the perfect case for this. You need them twice a year, always urgently, and they sit untagged in a folder because nobody tags a PDF for a problem they might have in 2027. Dishwasher flashing E24 mid-cycle and you don't remember the filename. Anyone actually solved this?

Six months ago I got tired of deciding, every single time, whether a document was safe to paste into someone else's model. So I've been building AutoBot: an open-source AI platform that runs on your own hardware. Voice, a real browser it can drive, search over your own files. Your data. Your AI.

Your documents. Your agent's memory. Every action it takes. All of it stays on your hardware — there's no "our server" to send it to. Open source. Your documents stay; your prompts travel only if you pick a hosted model. Your data. Your AI. github.com/mrveiss/AutoBot-AI

✨ This week in AutoBot: ✨ fix(provision): stamp node_roles_declared from t. Shipped 3 highlights, +1 contributors. All open source → https://github.com/mrveiss/AutoBot-AI/commits/Dev_new_gui?since=2026-08-12T09:00:02.662484Z

Most "AI" just chats. AutoBot acts. It talks back by voice. It drives a real browser — seeing, clicking, filling forms. It runs the task, not just describes it. Open source, self-hosted, on your hardware. Your data. Your AI. github.com/mrveiss/AutoBot-AI

✨ This week in AutoBot: ✨ refactor(agents): split memory-monitor into capt. Shipped 3 highlights, +2 contributors. All open source → https://github.com/mrveiss/AutoBot-AI/commits/Dev_new_gui?since=2026-08-05T09:00:03.299578Z

🚀 Two weeks of AutoBot: 231 fixes, 53 features, 16 perf wins — now on main. • Grounded /research: multi-source corroboration + contradiction surfacing • SSRF guard hardened across fetch and browser paths `main` is the default branch; Dev_new_gui gets it first. github.com/mrveiss/AutoBot-AI

"The cloud is just someone else's computer." So why does your AI agent run there? AutoBot runs on yours. Open source, self-hosted — voice, browser, real work. Your documents stay put; run a local model and nothing travels at all. Your data. Your AI. github.com/mrveiss/AutoBot-AI

✨ This week in AutoBot: ✨ test(kb): make the GPU chunker and KB stats test. Shipped 3 highlights, +2 contributors. All open source → https://github.com/mrveiss/AutoBot-AI/commits/Dev_new_gui?since=2026-07-29T09:00:02.182745Z

Rented AI can change the rules overnight. Yours can't. No subscription. No surveillance. No vendor holding your keys. Deploy AutoBot once — one install script — and run it your way, forever. Your data. Your AI. github.com/mrveiss/AutoBot-AI

✨ This week in AutoBot: ✨ fix(analytics): exclude test modules from the en. Shipped 3 highlights, +2 contributors. Check it out → https://github.com/mrveiss/AutoBot-AI/commits/Dev_new_gui?since=2026-07-22T09:00:02.984622Z

Don't just chat with one AI. Build a workforce. Compose multi-step agents on a visual canvas. Add modules: an agent-company (AutoBot LLC), a transcriber, codebase analytics (early). All on infrastructure you own. Your data. Your AI. github.com/mrveiss/AutoBot-AI

You shouldn't have to ask permission to use your own AI. Your data never leaves your machine — whatever LLM powers it. No rate limits on your knowledge. No vendor changing the rules overnight. Deploy once. Run it your way. Forever. github.com/mrveiss/AutoBot-AI

✨ This week in AutoBot: ✨ fix(llm,npu): openrouter LLMResponse field names. Shipped 3 highlights, +2 contributors. Check it out → https://github.com/mrveiss/AutoBot-AI/commits/Dev_new_gui?since=2026-06-29T05:00:04.393955Z

The main feature isn't the chat. It's the knowledge base. Feed it your docs. Your codebase. Your business. It learns what you know. It stays where you are. An AI that's actually about you. github.com/mrveiss/AutoBot-AI

✨ This week in AutoBot: ✨ feat(ansible): pg_hba supports coexisting app us. Shipped 3 highlights, +2 contributors. Check it out → https://github.com/mrveiss/AutoBot-AI/commits/Dev_new_gui?since=2026-06-22T05:00:02.773725Z

WordPress gave everyone a website. AutoBot gives everyone an AI. Self-hosted. Open source. Plug in any LLM — Ollama, OpenAI, Claude, Gemini. Your data stays on your machine. The brain is your choice. The AI platform that belongs to you — not to us. github.com/mrveiss/AutoBot-AI

✨ This week in AutoBot: ✨ fix: U7 #9926 leftover discoveries — hook over-b. Shipped 3 highlights, +3 contributors. Check it out → https://github.com/mrveiss/AutoBot-AI/commits/Dev_new_gui?since=2026-06-15T05:00:04.195437Z

Your data. Your AI. Self-hosted AI that actually belongs to you. github.com/mrveiss/AutoBot-AI

✨ This week in AutoBot: ✨ feat(secrets): UnifiedSecretsService — envelope . Shipped 3 highlights, +2 contributors. Check it out → https://github.com/mrveiss/AutoBot-AI/commits/Dev_new_gui?since=2026-06-08T05:00:03.682819Z

Answering AutoBot production questions 🧵 Q: "Does AutoBot enforce Docker resource limits?" A: Yes! hardened compose includes: • Memory: 768M-1536M/service • CPU: 0.75 cores • PID limits: 128-256 (fork-bomb protection) • Read-only filesystems

✨ This week in AutoBot: ✨ feat(backend): lightweight inference mode — bypa. Shipped 3 highlights, +2 contributors. Check it out → https://github.com/mrveiss/AutoBot-AI/commits/Dev_new_gui?since=2026-05-25T05:00:04.234278Z

Your data never touches our servers. No cloud vendor to breach. No third party holding your keys. What stays on your machine stays yours — full stop. You control the perimeter. We just give you the tools. github.com/mrveiss/AutoBot-AI