Start Your Agent

Start an agent on this computer, choose its model, and reuse its saved settings on later starts.

lnk agent start

When your agent doesn't answer, or answers slowly, its log says why.

Check What Starting an Agent Needs

You need a Mac, or a Linux x86-64 computer with bubblewrap installed. lnk up adds the modules, and also adds the one for a model on your computer. Local inference uses that computer's resources. A hosted model needs a key from Anthropic, OpenAI, or OpenRouter, and the provider bills you.

Add Cloud Services for Your Agent

Cloud connects your own clouds, runs agents on boxes, backs up files to buckets, and provides networking. IAM (identity and access management) controls access to resources. Start on this computer, then add the services your agent needs.

Start Your Agent in the Background

Your agent keeps running in the background, also after a restart. Each harness has commands of its own, lnk agent harness ..., for your agent's harness or the one --harness names. The names are link, mini, openclaw, hermes, deepseek, claude, codex, or one you added.

lnk agent start
lnk agent start --model.default local/qwen3
ANTHROPIC_API_KEY=sk-ant-... lnk agent start

Answer the First-Start Questions

The first start installs the harness. Then it asks which model your agent thinks with, which folder the agent works in, and which of Local Link's tools the agent gets. Memory is picked unless you say otherwise. Web asks for a Brave Search or Tavily key for its search. Browser needs Chromium or Google Chrome on this computer. Desktop needs Linux's Xvfb. Later starts reuse what you chose.

Your agent receives the Link tool by default. The tool reads the offline manual and inspects enabled capabilities, so you can ask your agent what it can do. Declared changes ask through the sandbox before they run. Remove the tool with lnk agent tool remove link.

Add the Tools That Start Off

A new agent runs Link's spec: file is on, so when you ask, the agent changes its files, instructions, and skills. Scheduler, run, and code are off until you add each one with lnk agent tool add <name> or enable it in your spec. Add reusable task instructions through Skills.

Choose Tools Once per Agent

The first start asks about tools once per agent, whatever you answer. Run lnk agent tool add <tool> to add one later. Picking Memory offers an embedding model too. Without the model, memory searches by words only.

Start an Agent Without a Terminal

Without a terminal to ask in, the agent's folder is ~/Link/Agents/<name>/home. --run.files <dir> picks another. The start says so and names the command for the agent's tools, and a later start in a terminal asks.

Choose From the Models Offered

The first start offers a model already running on this computer. With none, an Apple Silicon Mac is offered one to install, a few GB in size. After that comes an API key from Anthropic, OpenAI, or OpenRouter.

Read What Every Start Shows

Every start ends with the table lnk agent show shows. The table says whether your agent runs, its model, its tools, where you text it, and where its files are. Running means the harness answers. A key the provider refuses shows at your first message.

Follow the Suggested Next Commands

The start suggests lnk agent chat first, which talks to your agent in this terminal. Then comes lnk agent connect telegram while no connection reaches your agent. Next comes lnk agent tool add run, which lets your agent run programs, until it has Run. If you removed the Link tool, lnk agent tool add link lets your agent read Link's docs and run Link's commands again. Last come lnk agent harness set <harness> for another harness and lnk agent model set <model> for another model.

lnk agent chat

Give a Hosted Model's Key

Give a hosted model's key when lnk asks for it, or through the provider's environment variable: ANTHROPIC_API_KEY, OPENAI_API_KEY, or OPENROUTER_API_KEY. Never give the key as a command-line argument, which other users of this machine can see. lnk agent start --help and lnk agent model set --help say the same. To change the model later, see Switch Models. Your agent answers only while its machine is on.

Start an Agent Someone Made

An agent someone in your community made travels as one archive file, and so does one you made yourself. Share an Agent says how to export yours and import theirs, and what the agent needs before it starts.

Manage the Agent

Give each agent its own name and settings with Create Another Agent. Inspect an agent's configuration with Show an Agent, and diagnose runtime state with Status and Logs. Control the agent's service with Stop and Restart, and remove the agent with Delete Your Agent. Share an Agent packages reusable agents. Security and Decisions explain the lifecycle boundaries.