Add agents, adjust model behavior, and configure routing from one screen.
Multi-Agent OpenClaw
Multiple OpenClaw agents, one managed instance
Create separate OpenClaw agents for different roles without managing separate installs. Use each agent directly in Built-In Chat, or route Telegram and Slack conversations to the right one from a single hosted dashboard.
If one general-purpose agent is starting to mix your work, support, research, and operations context together, this gives you cleaner separation without the overhead of running multiple full environments.
Built-In Chat first
Pick the agent you want in chat and start working immediately. External routing is optional.
Per-agent separation
Each agent keeps its own workspace, sessions, role, and optional default model.
One managed runtime
Keep one hosted instance, create multiple specialist agents, and manage routing from one dashboard.
Why this matters
Why one agent stops being enough
Context starts colliding
One agent ends up holding founder tasks, support requests, research notes, and operational instructions all at once. The result is more confusion, weaker role discipline, and worse responses.
Separate installs are overkill
Running a whole new environment for every specialist role is expensive in attention. Most teams want role separation without multiplying hosting, setup, and maintenance overhead.
External routing gets messy fast
When Telegram and Slack conversations should reach different specialists, hand-editing config becomes a tax. A dashboard flow is easier to reason about and easier to keep correct.
What you can do
One hosted instance, multiple specialist agents
This is the right setup when you want one managed OpenClaw environment but more than one role inside it. Typical patterns include founder assistant + research agent, support triage + ops helper, or personal + work lanes that should not constantly mix context. If you are still deciding whether managed hosting is the right entry point, start with the main OpenClaw setup guide first.
- Separate roles: give each agent a focused responsibility instead of forcing one agent to do everything.
- Cleaner conversations: each agent keeps its own session history, so support threads do not pollute research work and vice versa.
- Model choice by role: override the default model for a specific agent when one workflow needs a different tradeoff.
- Less operational sprawl: keep one managed instance instead of juggling multiple partial setups.
Practical win
The value is not “more agents” by itself. The value is clearer role separation, easier routing, and less mental overhead while still keeping the managed-hosting simplicity that brought you here in the first place.
Fastest path to value
Use separate agents directly in Built-In Chat
The best part is that you do not need to start with Telegram or Slack to benefit from multiple agents. In OpenClaw Setup, you can choose the agent you want directly in Built-In Chat and begin using that role immediately.
- Pick the specialist you want in the chat UI.
- Test and refine each agent before exposing it to external channels.
- Keep external routing as a second step, not a prerequisite.
Built-in Chat support
In OpenClaw Setup, channel binding is optional in order to start playing with multi-agent OpenClaw. Once you configured the agents, you can message them right away using our Built-In Chat.
External routing
Route Telegram and Slack conversations to the right agent
Once the roles are working in Built-In Chat, you can wire external conversations into them. That means certain Telegram or Slack conversations can go directly to the right specialist instead of always falling into one default lane.
- Telegram direct and group conversations can be routed to a specific agent.
- Slack direct and channel conversations can be routed to a specific agent.
mainstays available as the fallback when no explicit route matches.
Routing in plain language
You decide which conversations belong to which specialist. The dashboard stores that mapping for you, so you do not need to hand-edit bindings every time you change structure.
How it works
What belongs to each agent vs what stays shared
Separate per agent
- Workspace, memory, skills and agent's markdown files
- Session history
- Name and description
- Optional default-model override
- Agent-local auth profile storage
Shared per instance
- Hosted runtime and infrastructure
- Billing and plan controls
- Globally configured skills
- Some provider setup in the current version
Why this split exists
You get meaningful role separation where it helps behavior and context, without paying the full operational cost of turning every specialist into its own standalone environment.
How this works in OpenClaw Setup
Agents configuration in OpenClaw Setup
Multi-agent setup is easily configured in OpenClaw Setup via the Agents tab — add new agents, customize their behavior, and route conversations from one screen. Screenshots below.
Decision guide
When to use multiple agents vs separate instances
Use multiple agents in one instance when…
- You want role separation inside one managed environment.
- You want to switch specialists quickly in Built-In Chat.
- You want Telegram or Slack routing without more hosting sprawl.
- You want one operational surface for one team or one owner.
Use separate instances when…
- You need fully separate runtime boundaries.
- You want independent instance-level credentials and controls.
- You want separate billing, ownership, or operational lifecycle.
- You are running materially different security or environment setups.
FAQ
Common questions about multiple OpenClaw agents
Can I use multiple OpenClaw agents without Telegram or Slack?
Yes. Built-In Chat is the simplest starting point. Just choose the agent you want and start working.
Does each agent get its own workspace and sessions?
Yes. That is the point of the feature: keep role-specific context and history cleaner inside one managed instance.
What is shared across agents?
The hosted environment, billing, and some provider setup remain shared per instance in the current version.
What happens if a conversation does not match a specialist route?
The main agent remains the fallback, so you always keep a default lane.
Ready
Launch OpenClaw multi-agent hosting without hand-editing config
Use separate agents from Built-In Chat first, then add Telegram or Slack routing when you need it.