You want an AI agent team in Discord — not one bot, but three, each with a distinct role, tagging teammates when a question crosses domains. The problem: OpenClaw gives you a single instance. Scaling to three means tripling the config, managing ports, and praying nothing conflicts.
ClawFleet manages multiple isolated OpenClaw instances from a browser dashboard with built-in team awareness. This tutorial: install, create three bots, give each a character, connect them to Discord, watch them collaborate. About 15 minutes on a Mac or Linux machine with 16 GB RAM.
Step 0: Install ClawFleet
One command handles everything — Docker, the CLI binary, the sandbox image, and the dashboard:
curl -fsSL https://clawfleet.io/install.sh | sh
For the full install walkthrough and troubleshooting, see: Self-Host OpenClaw on Your Mac in 10 Minutes. After install, the dashboard opens at http://localhost:8080.
Step 1: Register a Model
Go to Assets > Models and click Create. Pick any supported provider — OpenAI, Anthropic, Google, or DeepSeek. Paste your API key, select a model, and hit Test. ClawFleet validates the key before saving, so you'll know immediately if something's wrong.
One model is enough. All three bots can share it — models are reusable across instances.
Step 2: Create Three Instances
Go to Fleet and click Create three times. Give them meaningful names — I use engineer, researcher, and writer. Each instance is a fully isolated Docker container with its own filesystem and OpenClaw installation.
On a 16 GB Mac, three idle instances use roughly 4.5 GB RAM total. You'll barely notice them alongside your daily work.
Step 3: Give Each Bot a Character
Go to Assets > Characters and create three characters. Each has a Name, Bio, Communication Style, and Topics. Here's a team that works well together:
| Character | Bio | Style |
|---|---|---|
| Ada | Senior backend engineer. Go, Docker, system design. | Direct, code-first, uses examples |
| Max | Research analyst. Finds data, checks claims, cites sources. | Thorough, structured, evidence-based |
| Sam | Technical writer. Turns complex topics into clear docs. | Concise, reader-focused, avoids jargon |
Assign each character to its instance via the instance settings. ClawFleet renders these fields into a SOUL.md file inside each container. The OpenClaw gateway watches this file and hot-reloads on any change — no restart needed.
Step 4: Connect All Three to Discord
You need three Discord bot tokens (one per instance — channel assets are exclusive). For each:
- Discord Developer Portal > new application > Bot > Reset Token > copy
- Enable Message Content Intent under Privileged Gateway Intents
-
OAuth2 > URL Generator >
botscope +Send Messages+Read Message History> invite to your server
In ClawFleet, go to Assets > Channels, create three Discord channel assets (one per token), and click Test on each. ClawFleet verifies the token against Discord's API before saving. Assign one channel to each instance.
Step 5: Start and Watch Them Collaborate
Assign the model and channel to each instance if you haven't already, then click Restart Bot on all three.
Here's the key feature: ClawFleet automatically injects team awareness into every bot. When you assign characters to instances, each bot's SOUL.md gets a "Your Team" section listing its teammates — their names, roles, and which channel they're on. When the fleet changes, every running bot's teammate info updates via hot-reload. No manual editing.
Drop a message in the Discord channel that crosses domains:
"I need a technical blog post explaining our new caching layer. What's the best approach?"
What happens:
- Ada (engineer) explains the caching architecture with code examples
- Ada @mentions Max because the post needs performance benchmarks
- Max (researcher) contributes data and cites relevant sources
- Max @mentions Sam for the actual writing
- Sam (writer) synthesizes everything into a draft outline
The bots follow built-in collaboration rules: they @mention teammates when expertise crosses domains, build on each other's points instead of repeating, and yield to the human when the key perspectives are covered.
Resource Usage
Tested on an M4 MacBook Air (16 GB RAM):
| Instances | RAM (idle) | RAM (active chat) |
|---|---|---|
| 1 | ~1.5 GB | ~2 GB |
| 3 | ~4.5 GB | ~6 GB |
Three bots running alongside VS Code, a browser, and Slack — no noticeable slowdown.
What's Next
You now have a three-bot AI team collaborating in Discord — running on your hardware, data fully local, no subscription. From here: add more bots, install skills from ClawHub (13,000+ community), snapshot a well-tuned bot and clone it, or connect additional channels (Telegram, Slack, Lark). The wiki has guides for every provider and channel.
Get Involved
Star ClawFleet on GitHub — it helps others find the project.
Join the Discord — share your fleet setup, get help, and see what others are building.
Open source. MIT licensed. Free forever.
Top comments (0)