How multi-agent systems work in LimaCharlie

This video walks through how single agents and multi-agent systems are built and run inside the LimaCharlie platform.

Agents in LimaCharlie are defined declaratively. Each agent specifies the model it runs, its instructions, the tools it can access, what events trigger it, and the guardrails it operates under. This approach makes agents version controllable, reviewable, and portable across tenants.

Agents are triggered by platform events such as detections, case activity, scheduled intervals, or manual invocation. The triggering event provides the starting context the agent uses to begin reasoning.

Agents perform work through tools, and tool access is controlled through two layers:

• the tool allowlist configured on the agent
• the permissions associated with the API key used for execution

This structure keeps agent behavior bounded, observable, and auditable.

The video also demonstrates how multi-agent systems are orchestrated within the platform. Instead of relying on a single model to perform all tasks, LimaCharlie supports teams of specialized agents that collaborate through the case record.

A typical workflow includes:

• Triage agents that evaluate alerts and determine next steps
• Investigator agents that gather relevant context from telemetry
• Reporter agents that produce analyst-ready summaries
• Responder agents that take containment or remediation actions when appropriate

Each agent operates with its own prompt, tool access, and scoped permissions. Coordination between agents occurs through the case record, which acts as the shared context and provides a complete audit trail of the investigation.

The video also demonstrates how the AI Terminal can assist with scaffolding agents and multi-agent teams interactively inside the platform.

Learn more about LimaCharlie:
https://limacharlie.io

LimaCharlie provides a fully featured free tier with no credit card required.