ManageLM - The AI admin for your Linux and Windows fleet, on your LLM

I was looking for a AI shell / smart assistant for Linux and didn't find any satisfying solution; either not well designed or not smart enough for the admin tasks that I need, or simply too complex for very limited features. Finally all what I tested was far away from what I expected from a server assistant in 2026... So I decided to build ManageLM. The perfect AI tool for server management that I was was dreaming of. Tooks a few month of work and here it is. You work in plain English in a console or in full agentic mode with Claude, and an agent on the server does the job and reports. Two ways to work: - Assisted: a browser root console on any host, with an assistant that reads your terminal and proposes one command at a time. It has no shell of its own. You press the key. - Agentic: autonomous agents that plan and execute across the whole fleet. The part I care most about is that the AI never gets a blank cheque. Every command is validated against an allowlist written in code, not in a system prompt, and runs inside a kernel sandbox (Landlock/seccomp on Linux, a low-integrity process on Windows). Secrets are never shown to the model. Agents dial out, so there is no inbound port on your servers. And view-only AI is a per-server switch: run production read-only for reporting and incident work, keep staging fully agentic. You can also run the model yourself. Ollama, LM Studio, vLLM, llama.cpp, or cloud providers if you prefer. Point it at your own box and nothing your servers print ever leaves your network. There is even a No LLM mode that switches every model-backed feature off and leaves the deterministic half running: audits, compliance, monitoring, PKI, backups, credential rotation, consoles. It is one platform, not a chat layer on top of someone else's tooling: 33 skills (396 MCP operations), security audits with automated remediation, real-time threat detection, inventory, 13 compliance frameworks with auditor-ready evidence PDFs, 49 monitor types, PKI, encrypted backups, credential rotation, and a PKCS#11 keystore that holds your signing keys so your apps never do. It is also an MCP server, so Claude, ChatGPT, Cursor, VS Code, Goose, Slack and n8n all drive the same agents, with the same permissions and the same audit trail. Free forever for 10 servers. No credit card, no feature gates, self-hosted edition included. Built by RCDevs, where we have spent 15 years in cybersecurity, which is why "let an AI touch production" got argued about internally for a long time before any of this shipped. Whatever objection you have reading this, I have probably had it too. Ask me anything.