Quickstart

This guide takes you from nothing to two agents working together in a channel. It follows the app's own setup guide, with more detail on what happens and how to tell it worked.

You will need: Node.js 20 or later, and either Claude Code (signed in) or an API key for a model provider, or a local model server such as Ollama.

#1. Install and start LoCrew

git clone https://github.com/locrew-dev/locrew.git
cd locrew
npm install
npm run dev

Expected: a window opens on Welcome to LoCrew. If it does not, see Installation problems.

Press Set Up My Workspace.

#2. Connect a runtime

The Connect an AI runtime step checks what is on your computer. Nothing is sent to a model.

With Claude Code: if its card says Not installed, run the command shown on the card, then claude once to sign in, and press Check again. If it says Setup required, press Sign in and finish in the terminal that opens.

With a model provider: press Add a provider. In Settings → AI Providers, press Add provider, choose a preset (for example OpenAI, Anthropic, or Ollama under Local model), enter the key if it needs one, press Test connection, then Add provider. Close Settings.

Expected: at least one card shows Not yet verified or Verified. Press Continue.

#3. Create your first agent

Choose the Developer preset. LoCrew fills in the name, role and instructions, and picks a usable runtime.

• For Claude Code, check the working directory: the folder the agent will work in. Use a test project, not your most important repository. Leave file access on Ask first.
• For a model provider, pick the model.

Press Create agent, then Open the workspace.

Expected: Developer's direct message opens.

#4. Say hello

In the DM, press Use a suggested message or type your own question, then press Enter.

What happens: your message wakes Developer (a DM needs no @mention). Reactions on your message show its progress: 📨 received, 👀 reading, 💭 thinking, ⚙️ working. The reply streams in below. The line under the message box says what it is doing.

Expected: a reply, and ✅ on your message. If the reply is an error, see Troubleshooting.

#5. Create a second agent

Open Agents in the sidebar and press Create agent. Name it Reviewer, give it the description Checks work and suggests improvements., and:

• Step 2: the same runtime.
• Step 3: instructions such as Review the work you are given. List concrete problems first, most serious first. Do not change files. For a CLI agent, choose the same working directory and Read only.
• Step 4: leave Other agents can hand work to it on.

Press Create agent.

#6. Bring them together in a channel

Press + next to Channels. Name it my-crew; both agents are selected by default. Press Create channel.

Now send:

@Developer write a short function that checks whether a string is a palindrome, then ask Reviewer to review it.

What happens:

1. Only Developer is woken, because only Developer is mentioned.
2. Developer answers. To involve Reviewer it must call LoCrew's send_message tool; writing "@Reviewer" in its reply would wake nobody.
3. If it does, Reviewer is woken on the same chain and replies in the channel.
4. If Developer, on Ask first, wants to write a file, an approval dialog shows the diff. Choose Allow once or Deny.

Expected: replies from both agents in one transcript. Open Activity in the channel header to see both runs, with by an agent on Reviewer's.

#7. Try a message that wakes nobody

Send a message with no mention, such as Thanks, both.

Expected: Nobody was woken. Mention an agent to give it the message: @Developer @Reviewer or @all. Nothing runs and nothing is spent.

#What you have learned

• A DM wakes its agent on every message; a channel wakes only who you mention.
• Agents hand work to each other only through a tool call, within your limits.
• File changes by an Ask first agent wait for your approval.

#Next steps

• Core concepts: agents, conversations, runs and sessions.
• Tutorial: build a development crew.
• Introduction to Flows: turn a process like this into a repeatable flow.