Multi-agent · collaboration

Secure AI that works together

When agents delegate and share data, Burgus gives you visibility across the collaboration — who called whom, what moved where, and which alerts fired.

Visibility across the collaboration graph

When agents delegate tasks and pass data between services, security value comes from observing the mesh — not isolated hops. Burgus records per-request audit metadata, builds session and workflow timelines, and surfaces cross-agent policy hits and behavioral anomalies in one view.

Traceability modes

Default — route traffic through Burgus hostnames with registered keys. Per-request audit, inferred sessions, and temporal connection statistics — no agent changes required.

Enhanced — add optional X-Llmproxy-Session-Id, X-Llmproxy-Workflow-Id, X-Llmproxy-Agent-Id, and parent request headers for explicit multi-agent graphs and precise parent/child call chains. See integration.

Combined multi-agent security reports

Operator MCP and REST expose combined reports: traceability coverage, content-policy alert rollups, behavioral anomaly summaries, delegation fan-out, and burst timelines — one export for security engineers triaging multi-agent incidents. See Platform API.

Security alerts across the collaboration

• Cross-agent handoff and delegation abuse patterns
• Agentic abuse — jailbreak, tool abuse, coordinated fan-out
• Mapped to OWASP LLM Top 10 and MITRE ATLAS
• Active blocking on configured rules — credentials, injection, policy violations, and abuse patterns

Behavioral learning on collaboration

Automatic behavioral analysis learns normal collaboration patterns from your audit traffic — volume baselines, similarity bursts, graph topology signals — and flags deviations across agents. Behavioral analysis →