Install the mealissa-llm-agent

Install Ollama

Download and install Ollama for your platform. Ollama runs your models locally and serves them on localhost:11434. It must be running before you start the agent.

Pull a model

Pull any model from the Ollama library — Llama, Qwen, Gemma, Mistral, Phi, DeepSeek, and more:

ollama pull qwen2.5:4b

Confirm it's installed and Ollama is serving it:

ollama list

Get your license key

A valid license key is required to run the agent — it asks for the key the first time it starts (step 7), so have it at hand. Create an account and get your key from the Mealissa cloud console:

mealissa.com/cloud →  ·  See the licensing plans.

Download the agent

Grab the package for your platform from the downloads page. Each archive is around 440 KB.

Your browser will save the .zip to your Downloads folder.

Extract the archive

Unzip the downloaded package. Inside you'll find the agent launcher and a build folder. (There's also a bundled Java helper — you won't run it yourself; the agent calls it automatically only if needed.)

Run the agent

Launch the agent from the extracted folder. Make sure Ollama is running first.

mealissa-llm-agent.bat

./mealissa-llm-agent.sh

No need to set up Java yourself: on launch the agent checks for a compatible Java runtime (Java 11+) and installs one automatically only if it isn't already present. What to expect if that happens:

• Windows — you'll see an administrator (UAC) prompt, the installer downloads the Java runtime (~110 MB), and you'll then be asked to restart your computer. After restarting, run mealissa-llm-agent.bat again.
• Linux — the launcher asks for your sudo password and installs OpenJDK via your distribution's package manager (apt, dnf, or yum).

Answer the first-run prompts

On first launch the agent asks a few questions. Every prompt except the license key has a sensible default — press Enter to accept it. This is where you paste the license key from step 3:

╔═══════════════════════════════════════════╗
║  Mealissa Distributed LLM Local Agent     ║
╚═══════════════════════════════════════════╝

  License  []: Enter your license key: <your-license-key>
  Node ID  [4f3a1c2d-...]:
  Group    [9b2e0f1a-...]: lab
  Default  [N]: Is this the default server for the group? [y/N]: y
  Querying Ollama at http://localhost:11434... found 1 model(s).
  Available models:
    1. qwen2.5:4b  (4.0B params)
  Model    [1]:

  License  : ***XXXX
  Node ID  : 4f3a1c2d-...
  Group    : lab
  Default  : yes
  Model    : qwen2.5:4b  (4.0B params)
  Local    : 192.168.1.10:11434
  Server   : wss://mealissa.com/cloud/ws/node

• License — required. The agent encrypts it and saves it locally, so on later runs the prompt shows License [***1234]: and pressing Enter reuses it.
• Node ID — a unique identifier for this machine; the generated one is fine.
• Group — nodes in the same group can be queried together; the agent remembers your last group.
• Default — whether this node answers group queries by default.
• Model — pick a number from the models the agent detected in Ollama.

Verify your node

The agent connects to the Mealissa cloud over an encrypted channel and registers your node. Success looks like this in the console:

Connecting... (type 'help' for commands)
[connected] Registering node...
[registered] node_id=4f3a1c2d-...  public=203.0.113.5:54321

The prompt stays interactive — type status to see connection info, list to see other nodes in your group, or quit to disconnect. If the connection drops, the agent reconnects automatically.

Then open the cloud dashboard — your node and its model should appear online. For end-to-end proof, query your model from anywhere with the Anthropic-compatible API:

curl -s -X POST https://mealissa.com/cloud/g/<your-group>/v1/messages \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <your-license-key>" \
  -d '{"model":"qwen2.5:4b","messages":[{"role":"user","content":"Hello!"}]}'

Your local LLM is now globally accessible, encrypted in transit, with zero data stored.