Getting started
First launch and onboarding
What happens the first time you open LoCrew: the seven-step setup guide, connecting a runtime, creating your first agent, the Getting started checklist and the product tour.
The first time you open LoCrew on an empty workspace, a setup guide takes you from nothing to a working agent, a channel, a flow and an inspected run. It never calls a model on its own: the runtime checks read install and sign-in status only.
Who sees the guide
The guide appears automatically only when the workspace has no agents, channels, direct messages or flows. If you upgrade from a version that already has data, it stays out of your way. You can start it at any time from the workspace menu: Help → Restart setup.
#The seven steps
The setup guide has seven steps. The first three fill the window; the last four appear as a small Setup guide card on top of your workspace, with a highlight around the control to use.
| Step | Where | What you do | When it ticks off |
|---|---|---|---|
| 1. Welcome | Full window | Read the overview, choose how to continue | You press Set Up My Workspace |
| 2. Connect AI | Full window | Check Claude Code, Codex, a model provider or an external agent | A runtime is usable, or you continue without one |
| 3. Your first agent | Full window | Create an agent from a preset | The agent exists |
| 4. Say hello | Guide card | Send your first direct message | A message you wrote exists |
| 5. Create a channel | Guide card | Create #my-crew and choose who joins | A channel exists |
| 6. Your first flow | Guide card | Create and run a starter flow | A flow run completed |
| 7. Inspect a run | Guide card | Open a run and click a step | You opened a step's details |
Each step advances only when its milestone really happened. Skip step marks a step skipped; Back returns to the previous one.
#Step 1: Welcome
The welcome screen reads Welcome to LoCrew. with three cards: Conversations, Collaboration and Flows. You have three choices:
- Set Up My Workspace starts the guide at step 2.
- Explore First closes the setup and opens the Getting started checklist instead.
- Skip setup closes the setup and keeps the checklist closed.
You can come back from the workspace menu: Help → Restart setup.
#Step 2: Connect an AI runtime
This step checks what is already on your computer. Nothing is sent to a model. Each runtime has a card with a status badge:
| Card | What it is | What it needs |
|---|---|---|
| Claude Code | Anthropic's coding agent, running on this computer in a folder you choose | The Claude Code CLI and your Claude login |
| Codex | OpenAI's coding agent, running on this computer in a folder you choose | The Codex CLI (bundled with LoCrew) and your OpenAI login |
| Model provider | Any OpenAI-compatible API, Anthropic, Gemini, or a local model through Ollama | An API key (stored encrypted by your OS) or a local server |
| External agent | An agent that runs elsewhere and speaks the Agent2Agent protocol | Its Agent Card address; you add it when you create the agent |
The badges are Verified (an agent on it completed a real run), Not yet verified (installed and signed in, no run yet), Setup required, Verification failed, Not installed and Available.
What you can do on each card:
- Not installed: copy the install command shown on the card, install it in a terminal, then press Check again.
- Setup required or Verification failed: press Sign in. LoCrew opens a terminal running the CLI's login (
claude auth loginfor Claude Code;codex logout, thencodex loginfor Codex). Sign in there, then press Check again. - Model provider: press Add a provider to open Settings → AI Providers. See AI providers.
Press Continue when at least one runtime is usable. If none is, Continue without one skips the step; you can still create an agent, but it cannot answer until a runtime is connected.
Note
"Verified" means an agent really completed a run on that runtime. A CLI that is installed and signed in shows Not yet verified until its first real answer.
#Step 3: Create your first agent
An agent is a teammate with a name, a role and something powering it. The step offers three editable presets:
| Preset | Role | File access |
|---|---|---|
| Developer (default) | Helps write, explain, and review code. | Ask first |
| Researcher | Investigates topics and summarizes findings. | Read only |
| Reviewer | Checks work and suggests improvements. | Read only |
Fill in or adjust:
- Name (up to 64 characters) and Role.
- Look: a portrait.
- What powers it: Claude Code, Codex or Model provider. LoCrew preselects a usable CLI first, then a usable provider.
- For a CLI: the working directory and the file access level. Full access is never preselected.
- For a model provider: the provider and the model. Model agents are read only: they answer and can use tools you grant later, but never change files.
- Instructions (optional).
Press Create agent. LoCrew shows Created and then Open the workspace, which opens the agent's direct message.
To create an external (A2A) agent, use the full Create agent dialog instead; it needs the agent's card address and a connection test. See Creating an agent.
#Steps 4 to 7: in your workspace
Step 4, Say hello. A direct message is just you and the agent. Anything you send there starts a real run, with no @mention needed. Use a suggested message fills the box with Hi! Introduce yourself and tell me how you can help. so you can read, change and send it. The answer streams in below your message.
Step 5, Bring your crew together. Create #my-crew opens the real Create a channel dialog. You choose which of your agents join; no new agents are created. In a channel, type @name to choose who answers, or @all for everyone. A message that mentions no one wakes no one.
Step 6, Automate the work your crew does repeatedly. Create a starter flow creates My first flow: Start → HTTP request → Send message. It needs no AI: when you press Run it sends one request to api.github.com and posts what it found to your first channel. If an agent with a working runtime exists, **Or an AI example with agent** creates My first AI flow (Start → Agent), which is a real turn of that agent and may use its plan or API credit.
Step 7, Every run tells a story. Open the latest run, then click a step on the canvas to see its Input, Output, Events and Metrics. See Runs and debugging.
Warning
The starter flow requests LoCrew's GitHub repository. While the repository is private, that request returns 404 and the run fails. Use the AI example, or build your own flow.
When the guide finishes, a summary shows Your crew is ready. if an agent has completed a real run, or Your workspace is set up. with the items still open.
#The Getting started checklist
The Getting started button at the bottom of the sidebar shows your progress (N of 7 done). Each item ticks itself off when you have really done it:
- Connect an AI runtime
- Create your first agent
- Send your first message
- Create a channel
- Explore the Flow Builder (a flow exists)
- Run your first flow (a run completed)
- Inspect a flow run
Each open item has a button that takes you to the right place. Hide checklist hides it for good; you can reopen it from Help → Getting started or Settings → General → Open the checklist.
#The tour
Help → Take the tour walks through nine stops, each highlighting a real part of the window: navigation, Agents, Direct messages, Channels, Flows, the Flow Builder (only when a flow is open), Run history, Settings, and the Help menu. Use Next, Back, Skip tour or Esc.
#Tips
The first time you open a channel, the Flow Builder, a run, an agent's settings or an HTTP step, a one-time tip explains it. Got it dismisses one tip; Turn off tips stops them all. You can turn them back on under Settings → General → Show tips.
#Restarting the guide
Help → Restart setup (or Settings → General → Restart setup) returns to the welcome step. It keeps every agent, conversation and flow you have.
#Next steps
- Quickstart: the same journey with more detail on channels and hand-offs.
- Core concepts.