A local Manus alternative that codes, browses, and never phones home

What it does

AgenticSeek is an autonomous AI butler that breaks complex requests into steps and delegates them to specialized agents. It can browse the web hands-free through a headless browser, extract information, fill forms, write and debug code in languages like Python and Go, and read files from a local working directory. The whole stack—including SearxNG search and Redis—runs inside Docker on your own hardware, so your data stays off the internet.

The interesting bit

Rather than firing off monolithic cloud prompts, the system dynamically picks the best agent for each sub-task, like a tiny local team of experts. It is explicitly tuned for local reasoning models such as DeepSeek-R1 or Magistral, and bundles stealth-browser automation to dodge bot detection while scraping. The voice interface is still in progress, but the stated goal is a Jarvis-like assistant that lives in your basement, not a data center.

Key highlights

• Fully local execution via Ollama or LM Studio; cloud APIs are strictly optional.
• Multi-agent orchestration with automatic agent selection for browsing, coding, and planning.
• Headless web browsing using undetected Selenium and a bundled SearxNG instance.
• Autonomous coding and execution across Python, C, Go, Java, and more.
• Session memory and optional text-to-speech with multilingual support.

Caveats

• Voice input and the “Jarvis personality” mode are experimental and still in progress.
• The author describes it as an unfunded side-project with zero roadmap; durability is uncertain.
• A GPU capable of running 14B-parameter models is the minimum hardware bar.
• The config.ini parser cannot handle comments, which is a small but authentic rough edge.

Verdict

Developers with a decent local GPU who want a private, self-hosted alternative to cloud agent platforms should take a look. If you need polished enterprise reliability or are running on lightweight hardware, this is not your tool yet.