3rd Party Agents

Third-Party Agents

Third-party agents are AI systems you do not build or fully control — SaaS AI products, managed agent platforms, or agents running in environments where inline instrumentation or gateway proxying is not possible. For these, Highflame integrates at the network layer via Tailscale Aperture.

When LLM requests flow through Aperture, Highflame can evaluate them through two hook paths. The synchronous pre_request path calls Highflame before the request reaches the upstream model provider and can return an allow/block decision. The asynchronous tool_call_entire_request and entire_request paths send completed request context for tool-call visibility, sessions, and audit telemetry.

Key characteristics:

• No agent-side changes required — Aperture captures traffic at the network layer

• Synchronous guardrails — Highflame can allow or block pre-request traffic before it reaches the model provider

• Asynchronous observability — Highflame can record completed requests, tool calls, users, and sessions after the provider response

• Policy-mode control — monitor-mode policies record would-block telemetry while enforce-mode policies can block

Start here: Tailscale

For setup and configuration: Setup Guide · Troubleshooting