Llanite installs local LLM coding agents
One command installs a private local AI stack: LLM, Ollama runtime, coding agent, tools, and permissions — all configured and ready to run. Swap any component, inspect every detail, remove everything cleanly.
How it works
From zero to running in minutes
Llanite is designed to get out of your way. Three commands cover the entire workflow.
Install Llanite
One npm command gets you the CLI globally. No config files, no setup wizard, no accounts.
$ npm install -g @llanite/cliInstall a stack
Pick a stack profile. Llanite pulls the model and wires up the runtime — no manual configuration.
$ llanite install local-coderRun it locally
Launch the stack. All inference runs on your machine. Nothing phones home.
$ llanite run local-coderWhy stacks exist
Local AI should not require a weekend of config archaeology
Running a local model is only one part of the job. The hard part is making the model, runtime, agent, tools, context budget, and access policy work together without surprises.
Setup drift
Local agents often fail because model IDs, provider URLs, chat templates, and config files do not line up. Llanite packages those choices into inspectable stacks.
Hardware fit
RAM, VRAM, unified memory, quantization, and context length all change what feels usable. Llanite surfaces fit before you pull a model.
Tool calling
A model can chat well and still struggle with tools. Stacks pair models with agent layers that are known to support the workflow.
Context pressure
Large context makes local runs slow fast. Stack defaults should set a practical budget instead of pushing every file into every turn.
Policy ownership
Different agent layers enforce access differently. Llanite labels whether policy is Llanite-enforced, OpenClaw-managed, or externally managed.
Fast iteration
When a stack is almost right, you should swap one layer instead of starting over. Clone a preset, change the model or agent, and keep moving.
The case for local AI
Are local LLMs the future?
Cloud AI is convenient until you care about privacy, cost, or control. Local models have crossed the quality threshold for most coding and productivity tasks — and the pace of improvement is accelerating faster than any cloud provider can keep up with. Llanite exists because we think the answer is yes.
Privacy
Your code, queries, and context never leave your machine. No cloud provider sees your work.
Cost
No API bills. No per-token pricing. No rate limits. Run inference as many times as you want.
Performance
Modern 7B models match GPT-3.5 on most coding tasks. 27B models push further every quarter.
Registry to runtime
A stack should feel like one package, not six setup guides.
Llanite turns registry metadata into a runnable local agent profile: model, runtime, agent layer, tools, permissions, and launch command all resolved before anything runs.
inspectable
Every profile is plain YAML.
local-first
Ollama and local model caches stay on device.
permissioned
Shell, file writes, network, and secrets are explicit.
Ollama
Local model server
Qwen Coder
7B coding model
Llanite Agent
Chat, tools, memory
Filesystem
Read/write with policy
Git
Diffs and status
Confirm Shell
Permission gates
local-coder
resolved manifest
Built different
Everything you need, nothing you don't
One command. Model, runtime, agent — all ready.
Llanite detects your hardware, picks the right model for your RAM, and pulls everything needed to run locally. No config files. No setup wizard. No accounts.
Stacks
Ready to run, ready to inspect
Prebuilt local setups with explicit models, runtimes, agent layers, policy ownership, and hardware requirements. Everything visible upfront.
Local Coder
The default local coding stack: Qwen 3.5 27B Coder, OpenClaw, and OpenClaw's broad tool surface.
Compact
Read-only stack for code review, audits, and PR prep. No file writes or shell.
16 GB Coder
A smaller coding preset for 16 GB machines: Qwen 3.5 9B with the Llanite agent.
Your first stack is one command away.
Install Llanite, inspect the registry, and launch a local AI environment with explicit tools and permission gates.
$ llanite fetch
registry updated
$ llanite inspect local-coder
qwen2.5-coder:7b · ollama
$ llanite run local-coder
shell confirm · secrets block