Give your agents a machine of their own.
Merlin turns any Linux box or Mac into the place your AI agents work: real shell, real files, real git, reachable from any browser. If it runs in a terminal, it runs here: Claude Code, OpenCode, Gemini, Codex. Bring your own subscriptions; no middleman.
The Shift
The way we build software changed. The AI writes most of the code now; you direct, review, schedule. Work that way long enough and the pattern shows up: an agent needs somewhere to run, a way to act on its own, and a memory.
The Three Pillars
Somewhere to run: a coding environment.
A way to act on its own: an orchestrator, firing jobs on a schedule or a webhook.
A memory: a living knowledge base you both keep.
Three pillars, one project: an agent operating environment.
The coding environment
Run Merlin on any machine and control it from any device. A terminal where your agents run, the filesystem, and git history, all in the browser. What SSH gives you, made mobile.
Speak Instead of Typing
A microphone button on every text input. Press it and your speech transcribes into the terminal or prompt via Whisper. Useful on a phone where typing is slow, or when your hands are busy.
File Browser
A file browser with full access to your machine. Navigate any directory, preview images, open 3D models. Everything your laptop has, in your phone's browser.
Diff Viewer
Browse your git history and inspect what the agent changed. Walk through commits and read colored diffs in the same browser tab as the terminal.
Agent orchestration
Your agents run without you, on the server. A job is a script or an agent. Some run on a schedule you set. Others wake when the world sends an event.
Run on a Schedule
Agents don't have to wait for a prompt. Pair an instruction with a schedule, and the agent fires on its own, on the server. Useful for nightly reports, overnight research, recurring code maintenance.
React to the World
Point a webhook at Merlin and an event starts a job. CI breaks, and an agent reads the log and tells you why. A server drops, and it connects and reports what it found. It reacts the moment it happens, not the next time you look.
A living knowledge base
A memory you and your agents both keep. Markdown notes in your repo that agents read, write, and curate on a schedule. Persistent context that grows on its own.
Markdown Notes Shared With Your Agent
Markdown notes in your project repo. You and the agent both read and write them, which gives the agent persistent memory across sessions. Pair them with a scheduled job and the notes become a living knowledge base. The agent enriches, links, and curates them on its own.
No Server, Use Ours
Merlin runs for free on any Linux box or Mac you have: laptop, Raspberry Pi, cheap VPS. Open source, no account required.
Don't have a machine? Merlin Cloud gives you an isolated sandbox per project: 30 GB disk, 2 CPU, 8 GB RAM, up to 5 in parallel, for €39/month. Each one comes with Claude Code, OpenCode, Gemini, Codex, and Merlin pre-installed.
Either way, Merlin Cloud provides the access layer: HTTPS tunnel for any-browser access, secure authentication, Whisper transcription pre-configured. Run on your own infrastructure or ours.
First two weeks are on me.
sign_upFinal Thought
There's a particular morning. You open the terminal and the night has already worked: the jobs ran, the notes grew, a fix waits for review. At 03:12 a production server went down; a webhook woke an agent, the agent brought it back and wrote down what happened. You read it over coffee. After that morning, you don't go back.