4 bots running
open source · built on Claude Code

Your Discord server.
Now staffed.

Multiple Claude Code bots, distinct roles, independent sessions. Install, configure, launch.

$ npx disclaw-team init click to copy
# general

How it runs

Your Anthropic subscription. No API keys.

disclaw-team runs on Claude Code channels — each bot is a full Claude Code session, not an API wrapper. You're using the same subscription you already have. No per-token billing, no API keys to manage, no usage surprises.

Claude Code channels Anthropic subscription No API keys No per-token billing Full Claude sessions

How it works

Install, open the dashboard, and configure everything from the browser.

1
Install and launch
terminal
$ npx disclaw-team init

disclaw-team installed
Dashboard running

Open http://localhost:5173 to set up your team.
2
Configure in the browser
localhost:5173/onboarding
Set up your team
Step 3 of 4 — Assign roles
bot-1
Tech Lead ▾
bot-2
Frontend Dev ▾
bot-3
Backend Dev ▾
bot-4
QA Engineer ▾
Launch Team →

The onboarding wizard walks you through Discord setup, bot tokens, and role assignment.

3
Manage, swap roles, switch teams
localhost:5173/teams
disclaw-team
◉ Dashboard
⚙ Teams
★ Roles
⏱ Schedules
Team Presets
👔 Executive
CEO · Researcher · Validator · Engineer
💻 Dev Team active
Tech Lead · Frontend · Backend · QA
🎨 Frontend
Creative Dir · UI Designer · Dev · Reviewer
📦 Product
PM · UX Researcher · Dev · Copywriter

0
Built-in Roles
0
Team Presets
0
Role Tiers
0
Command to Deploy

The full toolkit.

Everything you need to run a real AI operations team — not a chatbot.

🎭
22 Swappable Roles
Orchestrators, specialists, executors. Assign any role to any bot. Swap anytime — roles are a library, not locked in.
role library
🌐
Web Dashboard
Live status, role picker, onboarding wizard, schedules, per-bot profiles, and role editor. All visual.
localhost:5173
💬
Task Channels
Orchestrator creates Discord channels per task. All discourse lives there. Summaries surface back to general.
per-task threads
🔐
Permission System
Default mode is fully autonomous — use only on trusted servers. Safe mode (--safe) routes every action to your Discord DMs for approval.
human-in-loop
Schedules
Recurring tasks via cron with a visual builder. Prompts injected directly into bot sessions on trigger.
cron + visual
🧠
Model Per Role
Opus for strategic orchestration, Sonnet for depth, Haiku for speed. Reasoning effort configurable per role.
multi-model

Pre-built team presets.

Pick a preset and your bots get roles instantly. Six ready-to-run compositions, or define your own.

Executive
CEO · Researcher · Validator · Engineer
research → validate → present
Dev Team
Tech Lead · Frontend Dev · Backend Dev · QA
architect → build → review
Frontend
Creative Director · UI Designer · Frontend Dev · Reviewer
design → implement → review
Product
Product Manager · UX Researcher · Frontend Dev · Copywriter
analyze → design → write
Content
Editor-in-Chief · Writer · Researcher · SEO
research → write → verify
Research
Lead · 2× Researcher · Data Analyst · Reviewer
investigate → analyze → peer-review

22 roles, fully configurable.

Each role defines personality, instructions, engagement rules, and model config. Edit or create your own via the web dashboard or YAML.

Orchestrators
opus · thinking/high
CEO
Tech Lead
Editor-in-Chief
Creative Director
Product Manager
Project Manager
Specialists
sonnet · thinking/high
Researcher
Validator
Data Analyst
Code Reviewer
Competitive Analyst
Security Advisor
SEO Specialist
UI Designer
UX Researcher
Executors
sonnet · standard/medium
Software Engineer
Frontend Dev
Backend Dev
Content Writer
QA Engineer
DevOps
Copywriter

One agent,
many masks.

Real agents.
Real discord.

A single-session approach was explored and rejected. It collapses into one mind performing disagreement. Separate sessions mean each bot builds its own context, forms its own views, and can genuinely push back.

The orchestrator delegates via Discord @mentions — never spawns subagents for specialist work. Specialists spawn their own subagents for deep work. The architecture mirrors how real teams operate.

Independent
Each bot reasons on its own with separate context. A researcher can genuinely disagree with the orchestrator — not perform disagreement.
⚙️
Parallel
All bots run simultaneously. No blocking, no queuing. The orchestrator delegates and moves on while specialists work in parallel.
👁
Visible
All discourse happens in Discord. Every decision, every pushback, every revision. Humans watch, interject, and redirect in real time.

⚠ Security Notice

By default, disclaw-team runs Claude Code with --dangerously-skip-permissions. This means anyone who can message your Discord server can trigger commands on your machine — file access, shell execution, and more.

Only run on servers where you trust every member. Use safe mode (--safe) to require approval for each action via Discord DM, or run on an isolated machine.

Ready to staff your server?

Open source. MIT licensed. One command to deploy.

$ npx disclaw-team init click to copy