An AI that answers only you

Your own AI, on your computer, that you text from your phone. It answers only you, remembers what you tell it, and can only touch its own folders.

  • Text it from your phone. Telegram, Slack, Discord, WhatsApp or Signal, and it answers you and no one else (Connections).
  • Choose its harness and its model. Start with Link Harness, or use OpenClaw, Hermes Agent, DeepSeek Harness or any other (Switch Harnesses). Its model can run on your computer or be a provider's (Switch Models).
  • It remembers, and works while you're away. It searches its files and conversations (Memory) and the web (Web), and on Link Harness it schedules work for itself and messages the rest of its team (Scheduler, Make a team).
  • It's sandboxed. It changes only its own folders and reaches the internet through a filtering proxy, and you choose what more it may do (Permissions).
  • It can move. Run it on a machine in your own cloud, and bring it back (Move Machines).

Quickstart

lnk up agents                 # the harness and Link Harness plugins (add `models` for local models)
lnk agent start               # pick a model, then Link Harness starts in the background
lnk agent connect telegram    # text it from your phone (or slack, discord, whatsapp, signal)
lnk agent status              # running and answering?
lnk agent create work         # another agent, with its own files and bot

A harness has commands of its own, lnk agent harness ..., for your agent's harness or the one --harness names: link, openclaw, hermes, deepseek or one you added.

What it needs

You need a Mac, or a Linux x86-64 computer with bubblewrap installed. lnk up agents adds the plugins, and lnk up models adds the one for a model on your computer. That model is free. A hosted model needs a key from Anthropic, OpenAI or OpenRouter, and the provider bills you.

Start your agent

It keeps running in the background, also after a restart.

lnk agent start                            # asks which model, then runs it in the background
lnk agent start --model qwen3              # a model running on this computer
ANTHROPIC_API_KEY=sk-ant-... lnk agent start   # a hosted model

The first start installs the harness and asks which model your agent thinks with and which folder it works in. Later starts reuse what you chose. The models it offers, in order, are:

  1. A model already running on this computer, which is free and private.
  2. With none, on an Apple Silicon Mac, one to install, a few GB.
  3. An API key from Anthropic, OpenAI or OpenRouter.

Give a hosted model's key as the provider's variable, as above, rather than with --key, because other users of this machine can see command-line flags. To change the model later, see Switch Models. Your agent answers only while its machine is on.

See how it's doing

lnk agent status      # running and answering, with its model, files, phone and sandbox
lnk agent logs --follow   # its last lines since the last start, then new ones

status exits nonzero unless the agent answers. When its proxy has paused its wire, or every wire is stopped, it says so and why. On a box, where no notification shows, that's where you see it. For other programs, status --json and list --json are in contracts.

See what it's told

lnk agent instructions          # the prompt its model gets, exactly as it's sent
lnk agent instructions --json   # each section, where it came from, its size

Link Harness builds its prompt from sections: Link's base, where it is (its files folder, which of its folders it may change, its channels), its skills, then your standing instructions, instructions/instruction.md in its files, and the notes it keeps for itself with its memory, instructions/notes.md. Each message starts with the date, the time, the channel and who's speaking, never the prompt, so the prompt stays the same and the model's provider caches it. Its spec's [instructions] replaces any section with a file of yours, or leaves it out. A harness from outside Link has its own prompt, which this can't show.

Make a team

A team splits a job by role: each agent keeps its own model, channels and files, sends the others messages, and shares a folder with them.

lnk agent team new research --agents scout,writer,checker   # each writes the folder they share
lnk agent team add research checker --role "checks every claim" --read . --write notes
lnk agent team list                                          # who's on it, and what each reaches
lnk agent team remove research checker
lnk agent harness check --harness link --agent scout         # it reads its team's folder, no other

On Link Harness a member sends another a message with its scheduler, and the answer comes back as a message from that agent, never from you. The scheduler's caps stop two that keep answering each other. The team's folder is ~/Link/Teams/<team>/Shared, and each member reaches only what it's granted of it, and no other team's. The team also brings a memory over that folder, one server for all its members. Its file, ~/.config/lnk/teams/<team>.toml, holds the members and the opinions they share, and each change restarts the members that run. You talk to each member on its own channels.

Stop it

lnk agent stop        # `lnk agent start` starts it again, with the same settings

Remove an agent

lnk agent remove work                 # asks first, then whether its files go too
lnk agent remove work --yes           # without asking; its files are kept
lnk agent remove work --yes --files   # and its files go too

remove stops the agent and deletes its settings, its harnesses with their installs, memory and conversations, its log, its keys, its sandbox rules and its memory's and scheduler's tools. Its files folder goes only if you say so; kept, it's moved to ~/Link/History/Agents/<name>-<date> for 30 days, and a files folder you chose yourself stays where it is. Its buckets stay in your clouds, so lnk agent restore work brings it back, asking for their password. What lnk measure recorded of it stays until it ages out. With one agent left, commands act on it without --agent.

Where things live

Your agent's files, its harness and its log each have a folder of their own.

  • ~/Link/Agents/<agent>/Home holds your agent's files. lnk uninstall keeps them unless you pick agents.
  • ~/Link/Agents/<agent>/Harnesses/<name> is a harness's home, with its install, memory and conversations.
  • ~/.config/lnk/agents/<agent>/logs/ holds its log.

Everything else Link keeps for an agent, and who can reach it, is in Security.

Troubleshooting

SymptomFix
"the … installer Link downloaded isn't the one it pins"A bad download, often a rate limit. Nothing ran; try again later.
start offers no local modelslnk up models, then lnk model install or start your runtime.
Linux: "the sandbox needs bubblewrap"sudo apt install bubblewrap
Ubuntu: bubblewrap can't make the sandboxLet lnk agent start add the AppArmor profile.
Linux: the agent stops when you log outsudo loginctl enable-linger $USER
status says it isn't answeringlnk agent logs --follow, then the Link Harness page, or its harness's own page.
Getting hallucinations, or it forgets what was said earlierIts window may be full, and what left it summarized. A larger [window] max in its spec helps, at a higher price per message.

Every flag is in lnk agent <command> --help. Security has the guarantees and gaps, and Decisions why it works this way. To add a harness, see Plugins.