Manual installation

Mycel needs three things before it can start: a Go toolchain to build the project, Ollama to run the model(s), and Docker (i.e., Redis to store conversations, etc.). This page walks through them one by one, then builds and configures the agent.

1. Go

Mycel is built with Go — see go.mod for the version the module targets.

brew install go   # macOS, or download from go.dev/doc/install
go version

The binary lands in $GOPATH/bin (usually ~/go/bin), so make sure that directory is on your PATH:

export PATH="$PATH:$(go env GOPATH)/bin"

2. Ollama

The model runs locally through Ollama.

brew install ollama   # macOS, or download from ollama.com/download

You do not need to start it or pull the model yourself: on startup Mycel runs ollama serve if nothing is listening, then ollama pull <LLM_MODEL> and streams the download progress to the terminal. All it needs is the ollama binary on your PATH.

Pick the model with LLM_MODEL (see Configuration). Mycel inspects the model's capabilities at startup and warns you about what it cannot do — for example a model without vision support will reject images, and one without tools support will ignore the email tool.

Choosing a model

A model with tools, vision and thinking support gets you every feature. Smaller models work fine for plain conversation but often drop tool calling first.

3. Redis

Conversation history lives in Redis. The repo ships a docker-compose.yml with everything you need:

docker compose up -d redis

To also get Redis Commander, a web UI for inspecting what Mycel has stored:

docker compose --profile tools up -d redis-commander

Already have a Redis instance? Point REDIS_ADDR at it instead and skip Docker entirely.

4. Configure

Copy the sample config, then fill in your values — every variable is documented in the configuration reference:

cp .env.sample .env

5. Build the binary

go install compiles the agent and writes the mycel binary to $GOPATH/bin, the directory you put on your PATH in step 1:

go install ./cmd/mycel

Confirm your shell can find it:

which mycel   # → /Users/you/go/bin/mycel

6. Set config

Mycel reads a single fixed path ~/.config/mycel/.env. Copy your filled-in .env (or .env.sample) there:

mkdir -p ~/.config/mycel
cp .env ~/.config/mycel/.env

From now on, ~/.config/mycel/.env is the file to edit when you want to change how the installed agent runs. Variables already exported in your shell take precedence over it, so LLM_MODEL=llama3.2 mycel works as a one-off override.

You are done — run mycel from anywhere:

mycel

Optional toolchains

Only needed if you plan to work on Mycel itself:

• Docs — make docs builds MkDocs into its own virtualenv, so it needs Python 3.9+ with the venv and pip modules (on Debian and Ubuntu those live in the separate python3-venv and python3-pip packages).
• Lint — make lint needs golangci-lint: brew install golangci-lint.