Integration Guides
RADAR works with any agent framework, LLM gateway, or custom stack. It observes existing infrastructure without requiring code changes.
How Integration Works#
RADAR does not require SDK installation, agent rewrites, or framework-specific adapters. It connects beside your existing infrastructure by consuming telemetry from your LLM gateway, tool execution logs, or model response streams. Supported integration patterns:
Gateway tap
Connect to your LLM gateway (OpenAI, Azure, Anthropic, custom) via log ingestion or event stream. RADAR captures prompts, completions, and metadata without modifying the gateway configuration.
Sidecar proxy
Deploy RADAR as a sidecar that observes agent-to-LLM traffic. No changes to agent code. Supports OpenTelemetry and custom webhook formats.
API push
Send evidence events directly from your agent runtime via REST API. Useful when you want explicit control over what gets recorded and when.
LangChain#
LangChain is the most widely adopted agent framework. RADAR integrates via the LangChain callback system or by consuming gateway logs.
Gateway log ingestion (recommended)
If you route LangChain agents through an LLM gateway, configure RADAR to consume gateway logs. Zero agent code changes.
Supported gateways: OpenAI, Azure OpenAI, Anthropic, any gateway exposing OpenTelemetry traces or structured JSON logs.
Direct callback
For teams that want explicit trace boundaries, use the RADAR REST API to push structured events from LangChain callbacks.
LangGraph#
LangGraph adds stateful graph execution to LangChain. RADAR observes the full graph execution — node transitions, tool calls between nodes, and final outputs — through gateway telemetry.
Graph tracing
Every node execution, edge traversal, and state mutation is recorded as a trace event. Graphs with parallel branches produce concurrent trace segments.
Tool isolation
Tool calls made within graph nodes are captured independently. RADAR maps each tool call back to its parent node and the overall graph session.
CrewAI#
CrewAI orchestrates multiple AI agents as a crew. RADAR captures evidence per-agent within the crew and correlates activity across the full crew execution.
Per-agent traces
Each agent in the crew produces independent traces. RADAR correlates them under a single crew session ID for end-to-end audit.
Handoff recording
When agents hand off tasks to other agents, RADAR captures the handoff context, data passed, and authorization boundary crossed.
AutoGen#
AutoGen enables multi-agent conversations. RADAR records full agent-to-agent conversations, tool invocation chains, and termination conditions.
Integration is done via gateway log ingestion or direct API push. No AutoGen-specific plugin required.
Custom Agent Stacks#
For teams using custom agent frameworks or proprietary stacks, RADAR provides a simple REST API for pushing structured evidence events.
REST API
Send traces and findings to RADAR via POST endpoints. Supports JSON and structured formats with session correlation.
Webhook receivers
Configure your agent stack to forward events to RADAR webhook endpoints. Supports batch and real-time modes.
Log file ingestion
Point RADAR at your existing agent log files. It parses structured logs and extracts evidence records automatically.
OpenTelemetry
RADAR accepts OpenTelemetry traces from any agent runtime that emits OTel signals. Plug and play with existing instrumentation.