Back to docs

Llanite guide

Open source coding agent

A fully open source coding agent uses open-source tooling at every layer: an open-source CLI to manage stacks, an open-source runtime to serve the model, and open-weights models that anyone can download and run. Llanite, Ollama, and models like Qwen 2.5 Coder and Gemma 3 are all open source or open weights. The whole stack is inspectable, free to use, and takes three commands to get running.

What makes a coding agent open source

An open source coding agent has three open layers: the agent tooling (the CLI and agent loop), the model runtime (the server that runs inference), and the model itself. Llanite is open source on GitHub. Ollama is open source. The models Llanite ships with — Qwen 2.5 Coder, Qwen 3, Gemma 3 — are open-weights models, meaning the weights are publicly available and can be run locally without a licence or subscription.

The Llanite open source stack

A Llanite stack ties together three open components: the Llanite CLI (which manages stacks, tools, and the agent session), Ollama (which serves the model locally), and an open-weights model. The stack is defined as a YAML manifest in the open registry, so you can read exactly what it installs, fork it, and modify it. Nothing in the core workflow requires a proprietary tool or a paid subscription.

$ npm install -g @llanite/cli$ llanite stacks$ llanite inspect local-coder

Open-weights models for coding

Llanite's coding stacks use Qwen 2.5 Coder, one of the strongest open-weights code models available. The 7B and 14B variants run on 16 GB machines; the 32B variant runs well on 32 GB. Qwen 3 8B is a good general model for smaller machines. Gemma 3 9B is useful for private chat without tools. You can pull any of these directly via Ollama.

$ ollama pull qwen2.5-coder:14b$ ollama pull qwen3:8b

Setting up the full open source stack

Install Ollama and Llanite, then install a stack. The llanite install command pulls the open-weights model through Ollama automatically.

$ llanite install local-coder$ llanite run local-coder
  1. 1Install Ollama from ollama.com (open source)
  2. 2Start Ollama: ollama serve
  3. 3Install Llanite: npm install -g @llanite/cli (open source)
  4. 4Check prerequisites: llanite doctor
  5. 5Install a stack: llanite install local-coder
  6. 6Run the agent: llanite run local-coder

Swapping models and customising stacks

Every component of a stack can be swapped without reinstalling from scratch. Clone a stack to create a custom copy, then change the model, agent, or runtime to experiment with different configurations.

$ llanite clone local-coder my-coder$ llanite set my-coder model qwen2.5-coder:7b$ llanite doctor my-coder$ llanite run my-coder

Contributing to the registry

Stacks are defined as YAML files in the open Llanite registry on GitHub. If you build a useful local AI stack — a specific model, agent, and tool configuration that works well for a particular workflow — you can contribute it as a pull request. The minimum stack schema requires a model, runtime, agent, tool list, permissions, and hardware requirements.

$ # Contribute at: https://github.com/lachlanforgan/llanite

FAQ

Are all models in Llanite open source?

The models Llanite ships in its default stacks — Qwen 2.5 Coder, Qwen 3, Gemma 3 — are open-weights models publicly available through Ollama. You can also set any stack to use a different model available in Ollama, including fine-tuned or custom models.

Can I add my own model to Llanite?

Yes. Use llanite set <stack> model <model-id> to point a stack at any model available in your local Ollama instance, including custom or fine-tuned models you've added with ollama create.

Is Llanite free to use?

Yes. Llanite is free and open source. Ollama is also free. The open-weights models are free to download and run. There are no subscriptions, API costs, or usage limits for running local stacks.

How do I contribute a new stack to the registry?

Stacks are YAML files in the registry at github.com/lachlanforgan/llanite. Open a pull request with your stack definition. The minimum required fields are id, name, runtime, model, agent, tools, permissions, and hardware requirements.

Can I use Llanite with a locally fine-tuned model?

Yes. If you have fine-tuned a model and added it to Ollama with ollama create, you can point any Llanite stack at it using llanite set <stack> model <your-model-id>.

What licence is Llanite published under?

Check the current licence at github.com/lachlanforgan/llanite for the latest information.