Back to docs

Llanite guide

Private AI coding assistant

A private AI coding assistant runs the model on your own machine, so your code never leaves your system. Llanite makes setting one up as simple as three commands: install the CLI, install a stack, and run it. The model runs locally through Ollama, the agent session stays on your machine, and nothing is sent to an external API unless you explicitly enable a tool that makes network requests.

How a local setup keeps code private

When you run a Llanite stack, the model is served by Ollama on your machine at localhost. The agent loop runs locally, the tools access your filesystem and shell locally, and prompt data does not leave your system. There is no API key, no usage billing, and no remote server involved in the core workflow. The only network activity is the initial model download when you first install a stack.

What Llanite adds to a private setup

Llanite makes the private setup easy to install, inspect, and repeat. Instead of manually configuring Ollama, an agent tool, and a set of permissions, you install a single stack that has already made those choices. The stack manifest shows exactly what the agent can access before it runs — you can verify the tool permissions before installing anything.

$ llanite inspect local-coder

Setting up a private coding assistant

Install Ollama, then install Llanite via npm. Three commands get you to a running private agent session.

$ llanite install local-coder$ llanite run local-coder
  1. 1Install Ollama from ollama.com and start it: ollama serve
  2. 2Install Llanite: npm install -g @llanite/cli
  3. 3Check prerequisites: llanite doctor
  4. 4Install a coding stack: llanite install local-coder
  5. 5Start the session: llanite run local-coder

Checking network permissions before running

Llanite stack manifests declare network access explicitly. The default coding stacks have network set to deny, which means the agent cannot make external requests during a session. You can confirm this before installing by inspecting the stack, and check the active permissions during a session with /permissions.

$ llanite inspect local-coder$ # During a session: /permissions

Changing models and components

Swapping the model in your private stack is one command. You can try different Ollama models, change the agent layer, or fork a stack into a custom copy without rebuilding the setup from scratch.

$ llanite set local-coder model qwen3:14b$ llanite clone local-coder my-private-coder

Use cases for a private coding assistant

A local private coding assistant is useful for proprietary codebases where sending code to an external API is not acceptable, compliance-sensitive projects with data residency requirements, offline development without a reliable internet connection, and any workflow where you want full control over what the agent can see and do.

FAQ

Does Llanite send any data to external servers?

No. Llanite communicates with Ollama's local server on your machine. No prompt data, file content, or usage data is sent to external servers during an agent session. The only network activity is the initial model download when you run llanite install.

Can I use Llanite in an air-gapped environment?

Yes, once the model is downloaded. Run llanite install on a connected machine to cache the model via Ollama, then use the stack offline. After the initial install, llanite run does not require network access.

How do I verify that my stack has no network access?

Run llanite inspect <stack> before installing to read the full permission manifest. Look for the network field — deny means the agent cannot make external requests. During a session, type /permissions to see the active permission state.

Is a local coding assistant as capable as cloud tools like Copilot or Cursor?

Smaller local models are generally less capable than large cloud models on complex tasks. However, Llanite's agent is designed to maximise the capabilities of smaller local models through efficient tool use and context management. For many common coding tasks — explanation, refactoring, navigation, documentation — a well-configured local stack is highly effective.

Does Ollama send telemetry or usage data?

Ollama does not send model inputs or outputs externally. Check Ollama's current privacy policy at ollama.com for the latest information on what diagnostic data, if any, is collected.

Can I use Llanite for work on a proprietary codebase?

Yes. Because the model runs locally and no data leaves your machine, Llanite is suitable for proprietary and confidential codebases. Always verify your organisation's policies on AI tool usage before using any coding assistant on work projects.