Moonlet
Moonlet is an agentic IDE that builds itself. It can control, customize, and automate itself, laying the groundwork for your own software factory.
This package provides the npx -y -p @halcyonlabs/app bb-app launcher, bundled bb CLI entry, and
Node SDK export. Every surface — the web app, CLI, and HTTP API — is a
first-class way to drive Moonlet. Work runs in threads you can follow live, steer at
any point, or hand off to another agent.
Note: Moonlet is in active development. Workflows and surfaces are still evolving.
Quick Start
Moonlet runs from npm and orchestrates coding agents you already have installed.
Prerequisites
- Node.js 22.19, 24, or 26.
- Git.
- At least one supported agent provider: Claude Code, Codex, Cursor via ACP, Pi, or another ACP-compatible agent.
If you already use one of these providers, Moonlet will pick up your existing credentials. If you use multiple providers, you can mix and match per task.
Supported host environments
- macOS
- Linux
Windows via Ubuntu on WSL2
Run all bb commands inside WSL2, install Node.js, Git, and your provider CLIs
inside that WSL2 distro, and use Linux-style paths such as /home/me/repo or
/mnt/c/Users/me/repo.
Inside WSL2, use Linux paths rather than drive-letter or UNC paths. Repos
inside the WSL filesystem are recommended;
/mnt/c/... is intentionally supported so you can keep an existing Windows
checkout, but it is slower and less reliable for file watching.
Install and run
npx -y -p @halcyonlabs/app@latest bb-app
Then open: http://localhost:38886
To opt into the automated nightly channel:
npx -y -p @halcyonlabs/app@nightly bb-app
Nightly versions are built from main and may be unstable. The nightly
dist-tag moves independently of the stable latest tag.
npm 12 and later block dependency install scripts by default. Moonlet needs those
scripts to build its native add-ons (better-sqlite3, node-pty,
@parcel/watcher). Without them Moonlet stops at startup with
Could not locate the bindings file. If your npm version is 12 or later, allow
the scripts for the install:
npx --allow-scripts=better-sqlite3,node-pty,@parcel/watcher -y -p @halcyonlabs/app@latest bb-app
Or set the policy once for all global installs:
npm config set allow-scripts=better-sqlite3,node-pty,@parcel/watcher --location=user
npx -y -p @halcyonlabs/app@latest bb-app downloads the published @halcyonlabs/app package, starts the server and
local host daemon, and serves the web app. It stores Moonlet-managed state under
~/.bb/ by default. If either managed child process exits unexpectedly, the
launcher restarts that child without stopping the other one. Press Ctrl+C in
the terminal to stop both processes and exit with status 0.
If another live Moonlet owns the runtime record for the same data directory and a
healthy server answers on the configured local address, startup exits with
status 1 without starting either child process. The existing instance keeps
running. A runtime record without a healthy server does not prevent startup.
Application logs rotate in logs/server.N.log and logs/host-daemon.N.log
under the data directory, keeping the five newest files per service. Output that
bypasses the logger, such as startup errors and crashes, goes directly to
logs/server-stdio.log and logs/host-daemon-stdio.log. These files append
across restarts; when one is over 10 MB as a service starts, it is renamed to
-stdio.1.log, replacing the previous one. The launcher prints status and log
locations without forwarding service output to the terminal, so a stalled
terminal cannot block service logging. To follow everything with the default
data directory:
tail -F ~/.bb/logs/*.log
Launcher status output is plain when stdout is redirected, including in CI.
Set FORCE_COLOR=1 to request color or NO_COLOR=1 to disable it; NO_COLOR
takes precedence. In-place progress updates require a stdout TTY.
The same output capture applies to bb-server and bb-host-daemon.
To stop a Moonlet install that runs in another terminal or in the background:
npx -y -p @halcyonlabs/app bb-app stop
stop reads bb-app-runtime.json from the data directory, confirms that the
recorded process really is that launcher, then stops it. Pass --data-dir when
the Moonlet install you want to stop does not use the default ~/.bb/.
The bb command
To control Moonlet from outside Moonlet, open Settings → General → Command line and
choose Install next to Moonlet CLI. Threads in Moonlet already have bb; this
is for your own terminals, scripts, and agents running outside Moonlet, such as
Claude Code or Codex. Moonlet links a launcher from ~/.bb/bin/bb into a folder on
your PATH (~/.local/bin on Linux; /opt/homebrew/bin, /usr/local/bin, or
~/.local/bin on macOS). On Windows it adds ~/.bb/bin to your user PATH;
open a new terminal afterwards. Remove takes it off again. Moonlet leaves any
other bb you already have in place.
From a terminal, before bb is on your PATH:
npx -y -p @halcyonlabs/app bb settings cli-command install
Updating from the app
When Moonlet has an update, Settings → Updates shows an Update button (or run
bb updates app apply). Moonlet downloads the new version into
~/.bb/app-versions/, restarts into it, and reconnects the page. Moonlet does not
roll back: if the new version fails to start, run a newer release
(npx -y -p @halcyonlabs/app@latest bb-app) or fix the cause. Later npx -y -p @halcyonlabs/app bb-app runs use the newer
installed version; pass --bundled to run the copy npx downloaded instead.
Start Moonlet with --no-in-app-updates to turn in-app updates off.
After the server moves to another machine, the old data directory keeps
server-moved.json. bb-app there starts no server: it runs this computer's
host daemon against the new server address in config.json, restarting it when
it exits, and answers the old server port. API requests get 410 server_moved
with the new address; browser pages redirect to it for a direct address or link
to it for Moonlet Connect. While a move back to this computer is in progress
(server-import.json), it frees that port for the incoming server. When
server-moved.json goes away, after that move completes or bb server unlock,
it starts the server and co-located daemon again. bb-server exits with status
3 instead of starting, except for the incoming server of a move back.
From the app, add or open a project, start a thread, and choose the provider you want that thread to use.
CLI
The package also exposes the bb CLI for an already-running Moonlet server:
npx -y -p @halcyonlabs/app bb --help
The CLI uses the same BB_SERVER_URL and Moonlet config resolution as the SDK. When
unset, it targets the default local packaged server at
http://127.0.0.1:38886.
Scripting with the SDK
The package also exposes a Node SDK for scripts that drive an already-running Moonlet server:
import { BBSdk } from "bb-app";
const bb = new BBSdk();
const thread = await bb.threads.spawn({
projectId: "proj_personal",
environment: { type: "host", workspace: { type: "personal" } },
prompt: "Summarize my active bb work.",
});
await bb.threads.wait({ threadId: String(thread.id), status: "idle" });
console.log(await bb.threads.output({ threadId: String(thread.id) }));
new BBSdk() uses the same BB_SERVER_URL and Moonlet config resolution as the
CLI. Pass new BBSdk({ baseUrl: "http://host:38886" }) for remote or test
targets (see the remote-access note below). Scripts launched by Moonlet already receive BB_SERVER_URL and
BB_THREAD_ID in their environment.
Provider Credentials
Moonlet uses whichever providers you have configured. Common providers:
| Provider | Setup |
|---|---|
codex | Install the Codex CLI. Then run codex login or configure credentials per the Codex docs. |
claude-code | Install Claude Code and authenticate per its docs. |
cursor | Install Cursor's agent CLI (cursor-agent) and authenticate per Cursor's docs. |
pi | Install Pi with npm install -g @earendil-works/pi-coding-agent (0.84.0 or newer) and authenticate per its docs; Moonlet can run the install from Settings. |
opencode | Install opencode and authenticate per its docs. |
grok | Install Grok Build and authenticate with grok login or XAI_API_KEY. |
hermes-agent | Install Hermes Agent, configure credentials with hermes model, then verify ACP with hermes acp --check. |
Moonlet indexes the documented native skill roots for Codex, Claude Code, Pi,
Cursor, OpenCode, omp, Grok Build, and Hermes Agent. It includes user roots,
project roots, and compatibility roots such as .agents/skills. These skills
appear in the selected provider's / command menu. The Skills page and
bb skill list show native skills for Claude Code, Codex, and Cursor. Moonlet also
reads configured Pi, omp, Grok, and Hermes skill directories, plus enabled
provider plugin skills.
Moonlet reads Pi's global ~/.pi/agent files and each workspace's .pi files.
This includes settings, credentials, models, packages, extensions, skills,
prompts, themes, and context files. Pi extensions can add models and tools.
Moonlet loads project resources only after Pi's saved or global trust policy approves
the workspace. An unresolved ask decision stays untrusted because Moonlet has no Pi
trust prompt.
You can still use the Pi CLI and /login to create this configuration.
Custom ACP agents are configured through the ACP providers plugin's
customAgents setting: bb plugin config provider-acp set customAgents '[...]'. See the configuration docs for the optional modelCli and
reasoningCli or nativeReasoning reasoning settings. The optional
nativeSkillRoots field adds provider-native skills to the composer. Its
user paths resolve from the target host home directory. Its project paths
resolve from the selected workspace.
Top-level sharedSkillRoots uses the same user and project path format.
Moonlet lists these sources as read-only skills. Moonlet injects them into Codex, Claude,
Pi, and ACP threads. This permits one physical skill collection for Moonlet and a
standalone provider CLI.
Configuration
Use bb-app config for persistent non-secret package settings under
~/.bb/config.json:
npx -y -p @halcyonlabs/app bb-app config set BB_APP_URL https://<machine>.<tailnet>.ts.net
npx -y -p @halcyonlabs/app bb-app config list
npx -y -p @halcyonlabs/app bb-app config refresh
For remote access, use Moonlet Connect or publish the default loopback listener with
Tailscale Serve. Direct tailnet or LAN access to port 38886 requires the
explicit, security-sensitive --server-bind-host 0.0.0.0 compatibility option;
see the multiple-devices guide.
Use bb-app client ssh-target to configure local editor opens for remote
Moonlet servers under ~/.bb/client.json. The target is the value that works after
ssh, such as devbox or user@devbox:
npx -y -p @halcyonlabs/app bb-app client ssh-target set https://bb.example.test devbox --host-id host_abc
npx -y -p @halcyonlabs/app bb-app client ssh-target list
Use bb-app env for provider credentials under ~/.bb/env.json:
npx -y -p @halcyonlabs/app bb-app env set OPENAI_API_KEY <key>
npx -y -p @halcyonlabs/app bb-app env list
npx -y -p @halcyonlabs/app bb-app env unset OPENAI_API_KEY
env list redacts all values. Config and env writes ask a running local Moonlet
server to reload; if Moonlet is stopped, the values apply on the next start.
For all config keys, precedence, startup flags, and source-development .env
behavior, see the
configuration docs.