Invite-only alpha for Mac
Chat with your Claude agents. Answer them in one inbox.
Lou is a Mac app for working with Claude agents on your own repos. You talk to an agent in a conversation, as in a CLI session, and anything it needs from you lands in one inbox. Commands that leave your Mac or can't be undone wait for you, with the exact command on screen.
Free. Your agents run on your own Claude credentials, so Anthropic's pricing and limits apply.
How it works
- 1Add a repo.Point Lou at a project folder on your Mac, or start a new project. Chat with plain Claude or any agent in your ~/.claude/agents folder.
- 2Start a conversation.Pick an agent and talk to it, as you would in a CLI session. Agents you run in the CLI can file requests into Lou too. Each conversation has a spending cap and a Stop button.
- 3Answer in one place.When an agent needs you, a request lands in your inbox with who is asking and why. You answer and it carries on.
One inbox, six kinds of request
Agents stop for different reasons, so each request has its own answer screen: decide between options, approve, run a command, change a setting, check something, or tell someone. Every request says which agent is asking, what it suggests and why. Requests an agent is stuck on come first, then the rest by priority and age, across all your projects. Answer it and the agent picks up where it stopped.
- Decide
- Approve
- Run a command
- Change a setting
- Check something
- Tell someone
See the command before anything risky runs
Commands that leave your Mac or can't be undone wait for you. Read-only commands and your project's own local scripts run without asking, so you are not asked about every git status. A held request shows the exact command, the folder it runs in and the git state, with a badge when it deploys a branch, changes a database schema, writes live data or could lose data. For those you type a word to turn Run on. Output streams back while it runs, and the agent sees how it ended.
Conversations with limits
Replies stream in as the agent writes them, and you can type ahead while it works. Paste or drop in screenshots. An agent can brief other agents, and the same cap covers them. They report back to the agent that briefed them. Set the cap from $1 to $250 per conversation, and stop it whenever you like.
One conversation to lead the rest
Turn on a lead for a project and your first new Claude conversation becomes its lead. It briefs agents, hears their reports and shows each one as a line in its conversation. When an agent asks a question, the lead answers it only from what you have already said, shows you the words it used and lets you change the answer. Anything else comes to you as one decision. When the conversation gets long, start a fresh one: agents at work move over, and nothing is lost. It is optional, off by default and a switch for each project.
A board for each project
Each project has a Board tab with four columns: Ideas, Planned, In progress and Done. You can add cards, and so can your agents, who move them as they work. When an agent says a card is finished, it shows Ready for you and stays in progress until you press Mark done. If it isn't finished, press Not yet. Start a conversation on a card to hand it to the project's lead conversation, or to a new one if there is no lead. The message is filled in, and nothing is sent until you press Send. Export writes the board to docs/BOARD.md in your repo. Lou never commits it.
What to know before you ask
- Mac only.
- macOS, nothing for Windows or Linux.
- Setup is by hand for now.
- Lou is not packaged as a download yet. It runs from source on your Mac, needs Node, pnpm, Postgres and the Claude Code CLI, and opens in your browser at localhost. Safari gives you Touch ID and Add to Dock. Expect to be comfortable in the CLI.
- Alpha: use it on repos you can afford to lose.
- Expect rough edges and changes between versions.
- It runs agents on your real repos.
- They work with automatic permissions. Lou asks you before commands that leave your Mac or can't be undone, and agents can't read your secrets files or change settings files. That protection has not had an independent security review, and commands like test and build run your project's own code, so treat Lou as a safety net, not a sandbox.
- Your Claude account, your usage.
- Lou runs agents with Claude credentials you provide, so Anthropic's pricing and limits apply to what they use. Lou shows a spending cap per conversation. It is an estimate, and a conversation can run slightly over it.
- Nothing goes to us.
- Lou runs on your Mac with a local database. You sign in with a token once, then with Touch ID. Your agents send what they read to Anthropic, as any Claude agent does; Lou has no servers that receive your code.
- The lead is off by default and has its own spending limit.
- You switch it on for a project in its settings. Its agents count toward one work limit, which you set there and can raise, and at the limit nothing new starts. The limit carries over when you start a fresh conversation.
- Support is as-is.
- You can report problems, but there is no promised reply time.
Request an invite
Tell us where to send it. Invites go out in small batches.
What happens next
- Your request
Your email and, if you like, a note.
- A place opens
Invites go out in small batches, so it may be a while.
- Your invite
We email you from hello@dfs.vc. Reply with your GitHub username and we give that account read access to the code.
Questions
What is Lou?
A Mac app for working with Claude agents on your repos. You talk to an agent in a conversation, and everything it needs from you (decisions, approvals, commands to run, settings to change, things to check) lands in one inbox. Each project also has a board, where you and your agents add cards and move them between columns, and you confirm what's finished.
What is the lead conversation?
An option for a project, off by default. When it is on, your first new Claude conversation becomes the project's lead. It briefs other agents and hears their reports. It answers their questions only in words you have already said, and shows you which words, with a way to change the answer. Anything else it passes on to you. When the conversation gets long, you can start a fresh one from the project's More menu. Nothing is lost.
Why an invite?
Lou is an alpha. We want a small group using it on real work first, so problems show up with people who can tell us. Wider access comes later.
Will it be open source?
Yes. The plan is a public repository under the MIT licence, with a date not set.
What does it cost?
Lou is free. Your agents run on your own Claude credentials, and Anthropic's pricing and limits apply to that usage.
Can I use my Claude subscription?
Setup details, including which Claude credentials the alpha accepts, come with your invite. Anthropic sets the rules for using a plan with other tools, so we won't guess here. The form asks how you pay for Claude so we can tell you what to expect.
Is it safe to run on my repos?
Use repos you can afford to lose. Lou asks before commands that leave your Mac or can't be undone, and blocks agents from reading secrets files or editing settings files. The same checks apply to agents your agent briefs. It has not had an independent security review, and it is not a sandbox: commands like test and build run your project's own code.
Does my code leave my Mac?
Lou has no servers and keeps its data in a local database. Your agents send what they read to Anthropic, the same as any Claude agent.
Does it replace Claude Code?
No. Lou runs alongside it. You talk to agents in Lou the way you would in a CLI session, and agents you run in Claude Code in the CLI can file requests into Lou after a one-time setup.
Which agents does it work with?
Claude agents. Other models are not supported.
What about Windows or Linux?
Not in this release.
What support will I get?
As-is. You can report problems, and there is no promised response time. Security reports go to hello@dfs.vc.
Is Lou made by Anthropic?
No. Lou is made by DFS and is not affiliated with or endorsed by Anthropic.