Multiple agents per project

A project can define multiple agents.

Each agent can have its own:

  • model
  • system prompt
  • tool set
  • role in your workflow

This lets you create setups such as:

  • a general-purpose coding agent
  • a reviewer agent
  • a docs agent
  • an orchestrator that delegates to specialists

Routing with @mentions

Type @ in the chat input to see the agents available in the current project.

Mentioning an agent routes your message to that specific agent.

Examples:

  • @docs rewrite the setup guide for clarity
  • @reviewer check this change for risky assumptions
  • @backend trace why this endpoint is timing out

Agent-to-agent delegation

Agents can also @mention each other.

This enables multi-step workflows where one agent coordinates or hands off work to another.

For example:

  1. an orchestrator breaks a task into parts
  2. a coding agent makes the change
  3. a reviewer agent audits the result
  4. a docs agent updates documentation

Model selection

Remote Lab supports multiple LLM providers.

A model can be selected globally for a conversation or overridden per agent. If an agent does not specify its own model, it falls back to the global model.

When to use multiple agents

Use multiple agents when you want clearer specialization, review boundaries, or delegation. For straightforward tasks, a single good default agent is often enough.