Automatic installation¶
Mycel needs three things before it can start: a Go toolchain to build the project, Ollama to run the model(s), and Docker (which runs Redis for conversations and SearXNG for web search). The installer puts all of them there for you.
Checking what is missing¶
To see the gaps without installing anything:
Installation¶
From a fresh clone, the following one command will install anything that's missing:
The installer inspects your machine, installs only the pieces that are absent, builds mycel, writes ~/.config/mycel/.env and pulls the model. It is safe to re-run, and it never overwrites an existing config.
Since installing Go, Docker or Ollama means running package managers and upstream install scripts, the run starts with a dry pass and shows what it is about to touch:
---> Looking at what this machine already has
ยท to install or start: ollama docker
? Go ahead? [Y/n]
There is nothing to confirm when nothing is missing.
Options¶
Pass these to ./install/install.sh directly:
| Flag | Effect |
|---|---|
--core-only |
only what is needed to run: skip the docs and lint toolchains |
--skip-model |
don't pull the model; Mycel pulls it on first run instead |
Supported platforms
macOS and Linux. On macOS the installer goes through Homebrew, offering to install it
if needed; on Linux it uses apt, dnf, yum, pacman or zypper, falling back to
the upstream installers for Go, Ollama and Docker. Windows works through WSL2.
Ready to run¶
After the installation is done, mycel is sitting in $GOPATH/bin (which is only added to your shell profile if it wasn't there previously), so you can launch the agent from any place on your filesystem.
The final step is up to you: provide values for your environment variables in ~/.config/mycel/.env (all variables are described in Configuration), then execute mycel.
Working on Mycel itself? Run make install again after a code change: it rebuilds and reinstalls the binary, and re-checking the dependencies costs well under a second when they are all already there.