Skip to content

rogerchappel/crewcmd

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

894 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

CrewCmd

Give your team AI superpowers.

Author's note: CrewCmd is still early. This README reflects what works today and the direction the product is heading, but some roadmap items are still in progress rather than fully shipped.

CrewCmd Introduction

CrewCmd is an Apache-2.0 open-source, self-hostable collaborative AI workspace built around channels and chats where humans and OpenClaw agents work together. Every person gets private AI leverage in their own runtime, while teams get shared channels with approved agents, tasks, inboxes, voice sessions, and governance.

Built for small teams that want to punch way above their weight without giving up personal runtime privacy.

A solo founder can keep private conversations with personal agents, then intentionally share selected outputs into a team channel. A 5-person startup can coordinate in shared marketing, product, or support channels with multiple humans, approved shared agents, voice chat, and approval gates without pretending Slack plus a bot is enough. That's the idea.

Private personal runtime · Shared AI channels · OpenClaw runtime integration · Agent inbox · Voice chat · Skills management · Task management · Agent team structure

CrewCmd Team Org Chart
Visual org chart for AI agents and delegation paths

Why CrewCmd?

Other platforms bolt AI onto an existing chat tool, or make every conversation feel like a private single-player assistant. CrewCmd is designed for the middle ground teams actually need: private personal AI work, shared collaborative channels, and a clear runtime boundary between the two.

  • Personal runtime privacy — Each user can have private OpenClaw-backed conversations, agents, tasks, and context that never leak into shared channels by default.
  • Shared AI channels when it matters — Bring humans and approved team/org agents into durable channels, project rooms, task threads, voice sessions, inbox items, and approvals when collaboration is useful.
  • Deploy a full team from a blueprint — Built-in blueprints currently cover the Solo Founder Kit, Startup Dev Squad, and Growth Team. Customize roles, hierarchy, and skills before deploying, or launch the blueprint as-is.
  • Bring your own OpenClaw agents — Import agents from an OpenClaw runtime and manage them in CrewCmd. Teams can bring their own personal OpenClaw agents, and companies can also provide shared or dedicated OpenClaw agents for team members.
  • One inbox for everything — Agents surface decisions, blockers, and completed work with priority tiers. No more checking 7 different tools. Review and approve from one place.
  • Skills management — Install, configure, and sync skills to your agents. CrewCmd includes built-in skills and can browse curated external skill sources, but the experience today is primarily about managing installed skills.
  • Channel and runtime access control — Personal agents stay private, personal runtimes are never attachable to shared channels, and company-owned agents can be shared with team or org visibility.
  • Humans and agents on the same board — Task management, project tracking, and time logging that works for both. See who's doing what, human or AI.
  • Agent team structure — Visual org chart for agents with reporting lines and delegation paths.
  • Governance foundation — Approval-gate and audit-trail primitives exist in the codebase, but the full guardrail workflow is still a work in progress.

CrewCmd Hands-Free Chat
Hands-free voice chat with your agents, from anywhere

Quick Start

The supported path for the controlled OSS preview is a source checkout with Node.js 22, pnpm 9.15, and embedded PGlite. It requires no external database.

Path Support level
Source checkout + embedded PGlite Supported for contributor evaluation and controlled OSS preview use
Docker Compose + Postgres Preview; complete the environment-specific checks in #668 before relying on it
External Postgres or platform deployment Preview; validate migrations, backups, TLS, and restore procedures in your environment
npm CLI, desktop packages, prebuilt server bundles, and published container images Deferred until versioned public artifacts exist

For normal local development on the machine running CrewCmd:

git clone https://github.com/rogerchappel/crewcmd.git
cd crewcmd
pnpm install
pnpm dev
# Open http://localhost:3000

Local network HTTPS

Use the self-signed HTTPS dev server only when you want to test from another device on your local network, such as a phone or tablet on the same Wi-Fi.

pnpm dev:https
# Open https://localhost:3000

The generated certificate is intended for local and LAN testing. It is not needed for Tailscale.

Tailscale testing

For Tailscale access, run the normal HTTP dev server and let Tailscale Serve provide the trusted HTTPS endpoint:

pnpm dev
tailscale serve http://localhost:3000

Then open CrewCmd from your device using the HTTPS URL shown by Tailscale, usually on your tailnet's *.ts.net address.

OpenClaw gateway setup

When importing your own self-hosted OpenClaw agents, use the gateway import flow in onboarding:

  1. Start CrewCmd using one of the local, LAN, or Tailscale options above.
  2. Choose Import existing OpenClaw crew during onboarding.
  3. Enter your OpenClaw gateway URL. Use localhost:18789 for same-machine testing, or the reachable host/IP/Tailscale name for a remote gateway.
  4. Enter the gateway auth token from ~/.openclaw/openclaw.json under gateway.auth.token.
  5. If pairing is required, run openclaw devices approve on the OpenClaw gateway host, then retry the connection.

Realtime voice through OpenClaw Talk

CrewCMD can opt into live, bidirectional OpenClaw Talk sessions while retaining the recorded speech-to-text/text-to-speech path as a fallback. Current CrewCMD realtime visuals require OpenClaw's gateway-relay transport.

  1. Configure a realtime provider under talk.realtime in OpenClaw with mode: "realtime", transport: "gateway-relay", and brain: "agent-consult". Keep the provider API key in OpenClaw; CrewCMD reads only the secret-free talk.catalog readiness response. See the OpenClaw Talk documentation.

  2. Confirm the selected gateway catalog without exposing provider secrets:

    openclaw gateway call talk.catalog --params '{}' --json

    The selected realtime provider must be configured and include gateway-relay in its transports. OpenClaw 2026.7.1 and newer also returns authoritative group readiness. CrewCMD remains compatible with 2026.6.11, but marks its older catalog as unverified until session creation.

  3. Set NEXT_PUBLIC_CREWCMD_REALTIME_VOICE=1 in .env.local (source/dev) or .env (Docker), then rebuild or restart CrewCMD. NEXT_PUBLIC_* values are embedded into the browser bundle and cannot be toggled after build time.

  4. Use localhost or HTTPS and allow microphone access. The voice UI reports disabled, provider-missing, unsupported-transport, unreachable, microphone-denied, and ready states before starting a realtime session.

If realtime readiness fails, CrewCMD labels the blocker and continues with the classic recorded STT/TTS path when that path is configured. Realtime provider credentials stay on the OpenClaw gateway.

Docker Compose

Docker Compose is intended for containerized self-hosting with a local Postgres container. This path is still less exercised than local dev, so treat it as preview until you have validated it in your environment.

cp .env.example .env
# Edit .env and set at least AUTH_SECRET.
# For Tailscale or another public URL, set NEXT_PUBLIC_APP_URL to that URL.
docker compose up

Compose reads .env by default. Use .env.local for non-Docker local development, and .env for Docker Compose unless you pass a different env file explicitly.

External Postgres

Use this for Neon, Supabase, or a self-hosted Postgres database:

cp .env.example .env.local
# Edit .env.local with your DATABASE_URL
pnpm install
pnpm db:migrate
pnpm dev

Requirements

  • Node.js 22+ and pnpm
  • Docker (optional, for containerized deployment)

Limitations

  • CrewCmd is still early software; verify auth, database, and OpenClaw gateway settings in a non-production environment before using it with a real team.
  • Docker Compose and LAN HTTPS flows are preview paths and should be validated on the target host and network.
  • No npm CLI, desktop package, prebuilt server archive, or published container image is part of the supported preview contract yet.
  • Governance and approval primitives exist, but broader guardrail workflows are still in progress and should not be treated as complete compliance controls.

Features

Feature Description
Private Personal Runtime Personal conversations, agents, tasks, and runtime-backed context stay private unless selected outputs are explicitly shared.
Shared AI Channels Team channels coordinate multiple humans and approved shared agents across conversations, task boards, inboxes, voice sessions, and approvals.
Team Blueprints Built-in agent team templates for solo-founder, development, and growth workflows, with optional customization before deployment.
OpenClaw Runtime Import Import and manage agents from an OpenClaw runtime, including runtime metadata, ownership, visibility, and sync.
Scope-Aware Inbox Centralized communication hub for visible decisions, blockers, approvals, mentions, and agent updates across private, channel, project, team, and org scopes.
Scope-Aware Skills Management Install, configure, and manage private, team, and org-approved skills, including curated browse/import flows and runtime sync.
Access Tiers Private personal agents plus team/org visibility for company-owned agents.
Scope-Aware Task Board Kanban and table views over every task the viewer can see, including private personal-agent tasks and shared channel/project/org tasks.
Team Structure Visual org chart for agents, with reporting lines and delegation paths.
Governance Foundations Approval gates, config versioning, and audit log primitives are present, with broader workflows still in progress.
Voice Chat Talk to your agents with speech-to-text and text-to-speech.
Light & Dark Themes Professional light theme for everyday use. Dark ops theme for power users.

Stack

  • Framework: Next.js 16 (App Router, Turbopack)
  • UI: React 19, Tailwind CSS 4
  • Database: PGlite (embedded, zero-config) or external Postgres
  • ORM: Drizzle
  • Auth: Auth.js / NextAuth v5 with credentials auth
  • Hosting: Self-hosted, Vercel, or Docker

Contributing

CrewCmd core is open source under the Apache License 2.0. Contributions welcome.

The project is designed to support both self-hosted deployments and a future managed CrewCmd service. The open-source core should remain useful on its own for teams that want to run agent operations on their own infrastructure.

git clone https://github.com/rogerchappel/crewcmd.git
cd crewcmd
pnpm install
pnpm dev

Open http://localhost:3000, create the first account, and run pnpm typecheck && pnpm test before opening a PR. Use pnpm dev:https only when you need HTTPS-only browser features such as microphone access.

Run the focused verification gates before shipping changes:

pnpm lint:check
pnpm typecheck
pnpm test
pnpm build
pnpm release:check

pnpm release:check is the canonical pre-release gate. It includes the app typecheck, unit tests, workspace smoke checks, npm package dry-run, and server release dry-run so CI verifies the same release surface maintainers run locally.

See CONTRIBUTING.md for the contributor workflow and docs/README.md for the documentation index.

License

Apache License 2.0 © 2026 RSCreative Technologies Pty Ltd.

About

Apache-2.0, self-hostable workspace for human teams and AI agents: channels, tasks, inbox, skills, governance, and OpenClaw runtimes.

Topics

Resources

License

Code of conduct

Contributing

Security policy

Stars

2 stars

Watchers

0 watching

Forks

Packages

 
 
 

Contributors